Large language model configured to direct domain-specific queries to domain-specific edge models
A hierarchical AI/ML assistant system with scale-specific LLMs and multimedia data processing addresses the challenges of petrophysical data complexity and connectivity limitations, facilitating efficient and accurate petrophysical analysis on edge devices.
Patent Information
- Application Number
- SA1120261747
- Authority / Receiving Office
- SA · SA
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-02-26
- Publication Date
- 2026-10-05
AI Technical Summary
The application of machine learning models in petrophysics faces challenges due to the complex and instrument-specific nature of petrophysical data, lack of focused training, and the need for real-time decision-making in environments with limited internet connectivity and stringent data security protocols, which hinder the deployment of large models like LLMs on edge devices.
A hierarchical AI/ML assistant system comprising a control LLM and scale-specific LLMs with reduced parameters, enabling efficient operation on edge devices by routing queries to specialized models tailored to petrophysical and field-specific knowledge, incorporating multimedia generation models for enhanced data processing.
Enables accurate, real-time petrophysical decision-making with reduced computing footprint, leveraging field-specific knowledge and multimedia data conversion for comprehensive petrophysical analysis in remote or disconnected environments.
Smart Images

Figure 09DE742D-D663-4601-9E48-C0DAB2CBE6EF 
Figure 1ECB2421-A3B7-46D6-930D-2F142E164560 
Figure 442ECF64-BCE5-46A9-8674-4F8522CFEF2D
Abstract
Description
Various aspects and examples of the Revelation are discussed in detail below. While specific applications are discussed, it should be understood that this is for illustrative purposes only. Those skilled in the relevant field will recognize the possibility of using other components and forms without straying from the scope and content of the Revelation. Therefore, the following descriptions and diagrams are illustrative and should not be interpreted as limiting. Many specific details have been described to provide a complete understanding of the Revelation. However, in certain cases, well-known or conventional details are omitted to avoid ambiguity. A reference to an incarnation or embodiment, an aspect or dimension, or an example or dimension in the current Revelation can refer to the same incarnation / epigram / dimensional aspect / etc., or any incarnation / epigram / dimensional aspect / etc., and such reference implies at least one of the incarnations, examples, and / or dimensions. Furthermore, the reference to "one of the incarnations" or "an incarnation" means that a specific attribute, structure, or property described in the incarnation is included in at least one incarnation mentioned in the revelation. The appearance of the phrase "in one of the incarnations" in different places in the description does not necessarily refer to the same incarnation, nor are they separate or alternative incarnations that preclude one another. Additionally, many attributes described are those that can be exhibited by some incarnations but not others. The terms used in this description generally have their usual meanings in the field, in the context of the disclosure, and in the specific context in which each term is used. Alternative language and synonyms may be used for any one or more of the terms discussed here, and no particular importance should be attached to whether a term has been described in detail or discussed in this document. In some cases, synonyms are provided for certain terms. The listing of one or more synonyms does not preclude the use of others. The use of examples anywhere in this description, including examples of any terms discussed here, is for illustrative purposes only and is not intended to further limit the scope and meaning of the disclosure or any illustrative term. Likewise, the disclosure is not limited to the various incarnations referred to in this description. Without intending to limit the scope of the disclosure, examples of measuring instruments, techniques, systems, devices, methods (also referred to here as processes), machine-readable non-temporal media, and their respective results are given below, in accordance with the examples and aspects contained in this disclosure. Note that headings or subheadings may be used in the examples to assist the reader, and these should in no way limit the scope of the disclosure. Unless otherwise specified, the technical and scientific terms used herein have the same meaning as generally known to persons of ordinary skill in the field to which the disclosure belongs. In case of conflict, the present document, including the definitions, shall prevail. The following description will present other features and benefits of detection, some of which will be partially evident from the description itself, or can be learned from applying the principles revealed here. The distinctive features and benefits of detection can be realized through the tools and combinations specifically described in the accompanying safeguards. These and other detection features will become clearer through the following description and the accompanying safeguards, or can be learned by applying the principles outlined here. Petrophysics is a branch of earth science that focuses on studying and understanding the physical properties of rocks and the fluids within them. Information about these physical properties is used in the oil and gas industry, as these properties can influence the location, size, and recoverability of hydrocarbons within a reservoir. Petrophysics can provide quantitative data that helps characterize the reservoir, assess its potential, and ultimately guide optimal extraction techniques. In petrophysics, various performance logging instruments are used to obtain different measurements of rock and fluid properties. Examples of these instruments include, but are not limited to, gamma ray, resistivity, neutron, density, acoustic, nuclear magnetic resonance (NMR), imaging logs, and drill cable formation testing devices. These logs provide measurements of various rock and fluid properties such as porosity, permeability, saturation, fluid type, and / or lithological properties, among others. However, applying a comprehensive petrophysical analysis presents a number of challenges. For example, measurements taken by performance logging instruments are often indirect, meaning there is inherent uncertainty in converting these measurements into meaningful reservoir properties. Furthermore, each performance logging tool responds differently to different formation conditions, and its measurements are often influenced by multiple variables. To improve the interpretation of these measurements, a deep understanding of the instrument physics and the application of experimental or analytical models may be necessary. Moreover, the reservoirs themselves are often complex, characterized by variations in lithology, fluid content, pressure, and / or temperature that can differ over small ranges. The resulting changes in petrophysical properties can further complicate interpretations. External factors such as drilling conditions and mud type can also affect log measurements and may require appropriate corrections. In addition, the high cost of well performance logging operations and the need for real-time or near-real-time decision-making increase the pressure to provide accurate interpretations quickly and efficiently. Furthermore, petrophysics is generally tied to the specific field in which it is applied. Each field presents a unique set of geological conditions, such as different rock types, depositional environments, fluid compositions, previous construction events, burial conditions, and so on. These variations can influence how different petrophysical properties are expressed and interpreted within the field. For example, the same performance log response might indicate the presence of highly porous sandstone in one field but low-porosity sandstone in another, depending on the lithological properties, fluid content, and reservoir condition. Therefore, knowledge of the specific field under study may be essential for accurate petrophysical interpretation. To this end, petrophysicists can incorporate field-specific knowledge into their interpretation models, making adjustments based on field observations, rock sample analysis, production data, and / or interpretations of previous well logs. To overcome these challenges, petrophysicists generally rely on a combination of specialized knowledge, practical experience, and advanced analytical techniques. The ultimate goal is to establish a comprehensive and accurate understanding of the reservoir that can support effective decision-making for exploration, drilling, and production activities. Each petrophysical problem can present its own set of challenges, often requiring a deep understanding of fundamental principles, strong problem-solving skills, and advanced reasoning abilities. In many cases, to address a problem, a petrophysicist may need to draw on a combination of petrophysical principles and a solution specifically tailored to the particularities of the scenario. Consequently, the expertise and experience of a petrophysicist can be invaluable in tackling such problems. Indeed, the ability to leverage prior knowledge while innovatively applying petrophysical concepts to new challenges can be crucial in the field of petrophysics.For example, a petrophysicist may need to critically evaluate data from different sources to identify patterns, establish connections, and make informed decisions under conditions of uncertainty. Many machine learning (ML) and artificial intelligence (AI) models, such as large language models (LLMs), possess remarkable capabilities in understanding complex knowledge domains, such as petrophysical principles, by training them on large datasets and implementing a vast number of parameters. These models can absorb and analyze enormous amounts of information, establish connections between them, and provide logical responses based on the data they have been trained on. Furthermore, the models can possess a degree of problem-solving skills. Given these capabilities, these models can serve as effective assistants to human petrophysicists. For example, considering an LLM's ability to recall petrophysical principles, examine data, and propose solutions based on a broad range of knowledge, an LLM can enhance the capabilities of a human petrophysicist. In some instances, LLMs can handle routine analysis, propose various solutions, and free up the petrophysicist to address other, more complex and ambiguous problems that might be better suited to a human approach. Despite the potential benefits of leveraging ML / AI models in petrophysics, significant challenges exist in applying these models to this field. For example, the complex and often instrument-specific nature of the data used to interpret petrophysical data can pose a major obstacle. Each performance recording instrument produces data that needs to be understood within its unique context, and each field of study has its own set of geological and physical considerations. Furthermore, complex field-specific knowledge, often acquired through years of experience and study, is frequently essential for petrophysical interpretations. While the versatility of models like LLMs trained on large datasets gives them a broad knowledge base, these models are not specifically trained as petrophysicists. Their understanding of the field may be drawn from a wide range of sources, which may not always reflect the depth and specificity of knowledge possessed by experienced petrophysicists. This lack of focused training can lead to gaps in their understanding and application of complex petrophysical concepts. Therefore, these models could greatly benefit from more specialized and integrated experience centered on petrophysical principles, tools, and field-specific case studies. This targeted approach can mimic the intensive training and practical experience that petrophysicists acquire over years, providing the models with a more nuanced understanding of the field, improving their ability to address specific petrophysical challenges, and increasing their potential as effective tools in the field of petrophysics. One of the major challenges to the practical application of petrophysics is that decision-making often takes place in locations with limited or no internet connectivity, such as at the well site or in remote field operations. Furthermore, due to the sensitive nature of oilfield data, many operations have stringent security protocols that mandate information storage on isolated servers or devices, prohibiting cloud computing. As such, a model like an LLM implemented as a petrophysics assistant may need to be configured to run on an edge device (e.g., at the site / field), eliminating the need for an internet connection or data transfer to remote systems (e.g., systems located away from the site / field). However, the sheer size of models like LLMs, which can include a very large number of parameters, presents a significant challenge to deploying them on local systems (e.g., edge devices).For example, edge devices may not be equipped with the processing / computational capacity required to run an LLM with a large number of parameters that have been adequately trained using petrophysical and field data. This section describes the key systems, devices, processes (also referred to as methods), and computer-readable media (collectively referred to as "systems and techniques") for implementing AI / ML that can be adequately trained using petrophysical and field data and can be implemented on the edge or any other system / device. In some examples, the AI / ML assistant may include an AI / ML model, such as an LLM, configured to route scope-specific queries to scope-specific models, which can be implemented on the edge (e.g., on the same local machine or another local machine(s)) and / or on any other system / device. In some respects, the AI / ML assistant's LLM can be trained on more general petrophysical and field data / knowledge, while scope-specific models can be trained more specifically and / or tailored to specific petrophysical or field data, topics, concepts, principles, and areas, etc.Furthermore, the AI / ML Assistant can address and overcome the above challenges by reducing the computing footprint / cost of the AI / ML Assistant when processing any given query(s) without reducing (or only slightly reducing) the overall capacity of the AI / ML Assistant compared to other AI / ML models such as other LLMs. In some examples, to address and overcome the challenges mentioned above, an AI / ML assistant could implement an architecture that includes a hierarchical system of scale-specific LLMs, each with fewer parameters and a smaller computing footprint than an LLM trained on multiple petrophysical scales. For example, an LLM with more general petrophysical knowledge and / or training could be configured to route scale-specific queries to scale-specific LLMs that are efficiently compressed to run on edge devices (and / or other devices) and have a reduced computing footprint, without sacrificing the depth and diversity of their knowledge and / or reasoning capabilities. Such an architecture could enable an AI / ML assistant to operate on an edge device for on-site use and could enable real-time (or near-real-time) petrophysical decision-making. An AI / ML assistant can be designed and trained in a way that respects the complexities mentioned above and incorporates instrument-specific and / or field-specific knowledge, resulting in an accurate and context-aware system capable of providing valuable insights and enhancing petrophysical decision-making. In some instances, an AI / ML assistant might include a petrophysical LLM configured as a control or management LLM with a larger number of parameters than the scalable LLMs that the control or management LLM queries for specific answers. In some cases, this control or management LLM could initially be trained on a broad / general dataset and optionally enhanced and / or retrained on petrophysical-specific data to increase its scalable knowledge and inference capabilities. Furthermore, smaller, scalable LLMs could be created, configured, and / or tailored to specific petrophysical topics, areas, principles, scenarios, use cases, and / or other specific aspects.In some instances, smaller, scope-specific LLMs can be trained on subject-specific data (e.g., codes) to create specialized models that focus on—or are adapted to—specific fields, areas, topics, principles, scenarios, or knowledge within petrophysics. Examples of such scope-specific LLMs include those trained and / or adapted to topics, areas, knowledge, fields, scenarios, use cases, and / or principles related to nuclear instruments, permeability, resistivity, continuity, stone properties, water saturation, pressure, fluid contact surfaces, formation testing, acoustics, electromagnetism, and / or porosity, among others. The LLM can be configured for control or management to act as an interface between petrophysicists and specialized LLMs (e.g., smaller, scaled-specific LLMs), and can be intelligently and / or selectively called upon to help resolve petrophysical queries. Thanks to the smaller sizes of specialized LLMs, these smaller LLMs can be efficiently run on edge devices, individually called upon by the LLM for control or management as needed to complete the task. This ensures a comprehensive and detailed knowledge base for real-time petrophysical analysis while minimizing the overall computing footprint of the system and complying with the often stringent requirements of field operations and data security. In some instances, a control or management LLM and / or any specialized LLMs can access a field-specific knowledge base and other petrophysical knowledge. In some cases, this knowledge base can serve as a sophisticated repository of results and similar findings from other petrophysical situations (e.g., similar situations) encountered in the field. By integrating information from this knowledge base, the control or management LLM can further refine its proposed solutions to petrophysical queries, enhancing accuracy and reliability. The AI / ML assistant can then leverage its inherent reasoning capabilities and its accumulated, field-specific knowledge to deliver a comprehensive and highly optimized solution. This dual approach ensures the AI / ML assistant's adaptability and robustness, enabling it to learn from past experiences while efficiently addressing new challenges. The petrophysical AI / ML assistant can optionally integrate a multimedia generation model, such as an image-to-text model, a video-to-text model, a speech-to-text model, and / or any other multimedia model. This multimedia generation model can convert graphs, maps, records, videos, images, diagrams, and / or other graphical information related to petrophysics into textual data that the LLM can process. In this way, complex visual data can be used effectively, ensuring that the nuances of context specific to each field are not lost or overlooked. Furthermore, this image-derived information can be used to improve the LLM and to implement specialized training for smaller, specialized LLMs. By enabling these models to learn from and adapt graphical data, as well as textual and other data, the AI / ML assistant's understanding of petrophysical domains can be enhanced. Moreover, this multifaceted learning approach can enable LLMs to handle the complex and diverse challenges of petrophysical interpretation more efficiently and accurately. Examples of the systems and techniques described here are illustrated in Figure 1A to Figure 8 and are described below. Figure 1A is a schematic diagram of an illustrative operating environment for performance-while-drilling (LWD) recording in a wellbore, based on some current exploration examples. The drilling rig shown in Figure 1A provides an example of a LWD configuration in a 100-wellbore scenario. An LWD configuration can include sensors (e.g., electromagnetic sensors, seismic sensors, gravity sensors, image sensors, etc.) that can acquire formation data, such as formation characteristics, formation components, and so on. For example, the rig shown in Figure 1A can be used to collect formation data via an electromagnetic imaging instrument (not shown) as part of an EMI (electromagnetic imaging) wellbore performance recording. The drilling rig in Figure 1A also illustrates what is referred to as measurement-while-drilling (MWD), which uses sensors to acquire data from which the wellbore trajectory and position can be determined in three-dimensional space. Figure 1A shows a drilling platform 102 equipped with a drilling turret 104 supporting a hoist 106 that raises and lowers a drill pipe string 108.The hoist 106 attaches an overhead motor 110 suitable for rotating and lowering the drill string 108 through the wellhead 112. A drill bit 114 can be connected to the lower end of the drill string 108. As the drill bit 114 rotates, it creates a borehole 116 that passes through various subsurface formations 118. A pump 120 circulates drilling fluid through a supply pipe 122 to the overhead motor 110, and down through the inner surface of the drill string 108 and out through ports in the drill bit 114 into the borehole. Drilling fluid returns to the surface via the annular space surrounding the drill string 108, and within the retention hole 124. Drilling fluid transports drilling outputs from wellbore 116 to the retention hole 124, and the presence of drilling fluid in the annular space helps to maintain the integrity of wellbore 116. Many materials can be used for drilling fluid, including oil-based fluids and water-based fluids. Performance logging tools (PLCs) can be integrated into the downhole assembly (DHA) ... One or more computing devices may be configured to control or monitor instrument performance, process performance logging data, and / or execute one or more aspects of the methods and operations included in the current disclosure. The downhole assembly 125 may also include a telemetry submersible 128 to transmit measurement data to a surface receiver 132 and to receive commands from the surface. In at least some cases, the telemetry submersible 128 connects to the surface receiver 132 by transmitting a wireless signal (for example, using mud pulse telemetry, electromagnetic telemetry, or acoustic telemetry). In other cases, one or more performance loggers 126 may connect to the surface receiver 132 by a wire, such as a wireline drill pipe. In some cases, the telemetry submersible 128 does not connect to the surface but instead stores performance log data for later retrieval at the surface when the performance log assembly is extracted. In at least some cases, one or more performance loggers 126 may receive electrical power from a wireline extending to the surface, including wirelines running through a wireline drill pipe. In other cases, power is supplied from one or more batteries or via a generator that is generated down the well. The 134 collar is a recurring component of the 108 drill pipe string and generally resembles a very thick-walled cylindrical tube, typically with threaded ends and a hollow core for conveying drilling fluid. Multiple 134 collars can be included in the 108 drill pipe string and are constructed and designed to be heavy to apply weight to the 114 drill bit, assisting in the drilling process. Due to the collar's thick wall, pocket-shaped or other types of openings can be provided in the collar wall without negatively affecting the integrity (strength, rigidity, etc.) of the collar as a component of the 108 drill pipe string. Figure 1b is a schematic diagram of a downhole environment with tubular elements, based on some examples from the current exploration. In this example, a 140 demonstration system is shown for downhole measurements after at least part of the wellbore has been drilled and the drill string removed. An electromagnetic imaging device (not shown) can be operated in the 140 demonstration system shown in Figure 1b to record the wellbore's performance. A downhole device containing a 146 instrument body is shown for performing performance recordings and / or other operations. For example, instead of using the drill pipe string 108 shown in Figure 1a to lower the tool down the well, which could contain sensors and / or other measuring devices to detect the properties and conditions near wellbore 116 and the surrounding formations and record their performance, a drill cable transport 144 could be used. The tool body 146 could be lowered into wellbore 116 by a drill cable transport 144. The drill cable transport 144 could be mounted on the drilling rig 142 or by portable means such as a truck 145.The transport system for a drilling cable (144) may include one or more wires, slip lines, cables, and / or similar components, as well as tubular transport systems such as coiled tubing, pipe fittings, or other tubular elements. The downhole tool may include an instrument used for collecting measurements during drilling, such as the electromagnetic imaging instruments described here. The 144-drill cable transport described provides power and support to the tool, as well as enabling communication between the 148A-N data processors located on the surface. In some instances, the 144-drill cable transport may include electrical and / or fiber optic cables for communication. The 144-drill cable transport is robust and flexible enough to connect the tool body 146 through the borehole 116, while also allowing communication through the 144-drill cable transport to one or more 148A-N processors, which may include local and / or remotely connected processors. The 148A-N processors may be integrated as part of a computer system, such as the computer hardware architectures described here. Furthermore, power can be supplied via the 144-drill cable transport to meet the tool's power requirements. For slipline or coiled tubing bodies, downhole power can be supplied using a battery or a downhole generator. Figure 2 is a diagram illustrating the architecture of the Petrophysical Assistant 200, based on some current discovery examples. The Petrophysical Assistant 200 can be used by any user, such as a petrophysicist(s), to obtain information such as answers to questions related to various issues, problems, properties, conditions, areas, topics, scenarios, fields, measurements, petrophysical contexts, and / or other aspects. The Petrophysical Assistant 200 can be configured to provide general and / or specific petrophysical information / knowledge in addition to field-specific information / knowledge. The Petrophysical Assistant 200 can have high overall capacity with a reduced computing footprint, allowing it to run on devices with fewer resources / capacities than servers or data centers, such as edge devices and client-architecture devices. The Petrophysical Assistant may include 200 Control LLMs and 210 Range-Specific LLMs. The Range-Specific LLMs may include 210 Specialist LLMs (e.g., specially designed, specially trained, tuned, etc.) for specific areas, topics, fields, scenarios, conditions, use cases, principles, domains, and contexts of Petrophysics, and / or any other aspects.For example, in the example shown in Figure 2, a specific range LLM could include a 210-trained and specially designed resistivity 212 model (e.g., customized for, modified for, etc.) to provide and / or embed knowledge about resistivity and / or inference capabilities; a 214-trained and specially designed nuclear magnetic resonance (NMR) model to provide and / or embed knowledge about NMR and / or inference capabilities; a 216-trained and specially designed porosity 216 model to provide and / or embed knowledge about porosity and / or inference capabilities; a 218-trained and specially designed continuity 218 model to provide and / or embed knowledge about continuity and / or inference capabilities; a 220-trained and specially designed fluid contact surface 220 model to provide and / or embed knowledge about fluid contact surfaces and / or inference capabilities; and a 222-trained permeability 222 model. Specifically designed to provide and / or incorporate knowledge of permeability and / or reasoning abilities, the Configuration Test 224 model is trained and specifically designed to provide and / or incorporate knowledge of configuration testing and / or reasoning abilities. The LLMs (Land-Specific) 210 in Figure 2 are merely illustrative examples provided for explanatory purposes.In other examples, the range-bound LLMs could include 210 other models not shown in Figure 2 and / or a different number of models than those shown in Figure 2. Each of the 210 scope-specific LLMs may be trained using one or more knowledge sources related to a particular scope of petrophysics (e.g., topic, area, field, domain, use case, scenario, condition, principle, context, concept, etc.), such as one or more scientific journal articles, scientific papers, field reports, question-and-answer pairs, petrophysics and / or wellbore records, instrument data, field data, training sessions / materials, tutorials, books, manuals, instrument data, data collection and processing, visual data (e.g., graphs, charts, maps, pictures, videos, records, etc.), publications, textbooks, internal knowledge materials, historical data, statistics, database knowledge, commentaries, research data, letters and / or notes from petrophysicists and / or other users such as other scientists, drilling reports, and / or any other petrophysics-related data and / or knowledge materials.In this way, each of the 210 range-specific LLMs can develop expertise in its own petrophysics range, allowing each of the 210 range-specific LLMs to answer queries related to its own petrophysics range / expertise. The LLM 202 can include a control LLM configured to route range-specific queries to specific LLMs from range-specific LLMs 210. This would allow the petrophysical assistant 200 to reduce its computational footprint when processing / answering any given query without reducing the overall amount of petrophysical and field-related knowledge and inference capabilities. For example, a particular petrophysical query might be best handled by one or more 210 range-bound LLMs, in which case it might be unnecessary and / or wasteful (for example, in terms of resources) to route the petrophysical query to all 210 range-bound LLMs, or to route the petrophysical query to a subset of 210 range-bound LLMs that include one or more range-bound LLMs that might be less appropriate to answer the petrophysical query, or to use a single LLM with the combined size and capabilities of all 210 range-bound LLMs to answer any given petrophysical query.Therefore, instead of directing a given query to all 210-range LLMs or to a single LLM that includes the parameters, knowledge, size, and / or overall inference capabilities of all 210-range LLMs, the 202-control LLM can intelligently determine which of the 210-range LLMs might be most appropriate to answer the given query, and selectively direct the given query to that LLM(s) for processing / answering. In the example above, since each individual LLM within the Specific Range 210 LLMs is smaller (e.g., in terms of parameters, overall size, etc.) and has a smaller computing footprint than all the Specific Range 210 LLMs, or than a single LLM with the same parameters, size, knowledge, and / or capabilities as all the Specific Range 210 LLMs, by intelligently routing queries to specific LLMs within the Specific Range 210 LLMs (e.g., instead of routing each query to all the Specific Range 210 LLMs or to an LLM with the same size, parameters, knowledge, and / or capabilities as all the Specific Range 210 LLMs), the Control LLM 202 can reduce the amount of resources (e.g., computing footprint, etc.) that the Petrophysical Assistant 200 uses to process / answer any query, which in turn allows the Petrophysical Assistant 200 to operate on machines with fewer resources and / or capabilities, such as machines or Client / edge architecture tools, etc.In this way, the Petrophysical Assistant 200 can be applied to any device and used by any user (e.g., a petrologist) in any field or environment, including fields or environments where access to the Internet and / or remote data is not available or permitted. The control LLM 202 can be trained on more general petrophysics and field data / knowledge compared to the scope-specific LLM 210, in order to gain a more general and / or broader understanding of petrophysics.For example, in some cases, LLM Control 202 may be trained using one or more sources of petrophysics and / or field knowledge, such as, for example, one or more scientific journal articles, scientific papers, field reports, question-and-answer pairs, petrophysics and / or wellbore records, instrument data, field data, training sessions / materials, tutorials, books, manuals, instrument data, data collection and processing, visual data (for example, graphs, charts, maps, pictures, videos, records, etc.), publications, textbooks, internal knowledge materials, historical data, statistics, database knowledge, commentaries, research data, letters and / or notes from petrophysicists and / or other users such as other scientists, drilling reports, and / or any other petrophysics-related data and / or knowledge materials.In some cases, the LLM 202 can be initially trained on a broad / general dataset, and can be optionally enhanced and / or retrained on petrophysics-specific data to increase its scope-specific knowledge and reasoning abilities. For example, in some cases, the LLM 202 can be trained using one or more knowledge / data sources to understand each of the domains of the Domain-Specific LLMs 210 in general, what questions might relate to each domain, what questions can (or might be best answered) by each of the Domain-Specific LLMs 210, which patterns suit a question that any of the Domain-Specific LLMs 210 might answer, what information (e.g., issues, features, conditions, principles, parameters, problems, domains, concepts, etc.) might relate to the domains (and / or questions associated with them) associated with the Domain-Specific LLMs 210, which model(s) among the Domain-Specific LLMs 210 might best understand a question and / or provide a more accurate and / or relevant response, and / or how to choose the model(s) from among the Domain-Specific LLMs 210. Intelligently directing any specific query to it.For example, the LLM 202 can be trained to understand that a particular question about the porosity of a particular configuration might be best answered by, for example, the resistivity model 212, the NMR model 214, the porosity model 216, and the configuration test model 224. The LLM 202 can then determine the resistivity model 212, the NMR model 214, the porosity model 216, and the configuration test model 224 to answer this query and route the query to the resistivity model 212, the NMR model 214, the porosity model 216, and the configuration test model 224. For example, and by analogy, a Control LLM 202 can function similarly to a primary care physician with general medical expertise, and a Scope-Specific LLM 210 can function similarly to specialist physicians with specialized expertise in one or more areas of medicine. In this analogy, a primary care physician can screen a patient for a specific condition and identify the most appropriate specialist physician(s) to assist with that condition. The primary care physician can then refer the patient to the specialist physician(s) for assistance with the specific condition (and / or consult the specialist physician(s) regarding the patient and the specific condition). Similarly, when a Control LLM 202 receives a petrographic query, it can identify the most appropriate model(s) from among the Scope-Specific LLMs 210 to answer the petrographic query and route the petrographic query to that model(s). LLM 202 can use any responses from this form(s) to provide a response to the query, as detailed here. In some examples, a control LLM 202 might include an LLM Interface 204 and a multi-pattern model 206. An LLM Interface 204 could include any software, interface, and / or component configured to act as an interface between a user(s) (e.g., petrophysicist(s), technician(s), engineer(s), etc.) and a range-specific LLM 210. For example, a user could access the petrophysics assistant 200 via the LLM Interface 204 and use the LLM Interface 204 to send one or more petrophysics queries and receive one or more responses from the petrophysics assistant 200. The LLM Interface 204 could communicate with any of the range-specific LLMs 210 to submit queries and receive responses from any of the range-specific LLMs 210. For example, the LLM Interface 204 could 204 To receive a specific query from the user. The control LLM 202 can intelligently determine which LLM(s) from the range-defined LLMs 210 can answer the query (and / or is the most appropriate to answer the query).The LLM Interface 204 can then selectively call / connect to the selected LLM(s) from among the range-defined LLMs 210 to obtain the relevant query response. The LLM Interface 204 can obtain the relevant query response from each LLM that has been queried. The LLM Interface 204 can then provide (e.g., display / present, supply, etc.) a response to the user. In some cases, before delivering a response to the user, the LLM 202 can modify the information obtained from each LLM that has been queryed. For example, the LLM 202 can combine, approximate, filter, rephrase / summarize, supplement, enhance, organize, structure, use, and / or otherwise refine any of the information obtained from each LLM that has been queryed to generate a response for the user, which can then be delivered to the user through the LLM 204 interface. For example, if the LLM 202 receives one or more query responses from one or more range-defined LLMs 210, the LLM 202 can use one or more of these query responses to generate a response to the query.In this example, the LLM 202 can combine one or more of the query responses or parts thereof to create a meaningful response, and can optionally use its broader and more general knowledge of petrophysics to create the response based on one or more of the query responses, make any changes to the response, add any relevant details, improve the dialogue that includes the response to the user, and / or edit or otherwise rephrase the response to the user. For example, if a query includes a question about the porosity associated with a particular configuration, the LLM 202 might determine that the resistivity model 212, the NMR model 214, the porosity model 216, and the configuration test model 224 are the most appropriate to answer the query and select these models to respond. The LLM 204 interface can then provide the query to the resistivity model 212, the NMR model 214, the porosity model 216, and the configuration test model 224 and receive a corresponding response from each model. The LLM 202 can then use the responses from the resistivity model 212, the NMR model 214, the porosity model 216, and the configuration test model 224 to formulate a response for the user. For example, the LLM 202 can combine responses (or parts thereof) from the resistivity model 212, the NMR model 214, the porosity model 216, and the configuration testing model 224 in an intelligent / feasible, systematic, and / or structured manner using the petrophysical knowledge and capabilities of the LLM 202.In some cases, the LLM 202 may use its knowledge and capabilities in petrophysics to improve or reformulate the resulting response based on responses from the resistivity model 212, the NMR model 214, the porosity model 216, the configuration testing model 224, and / or add any relevant information determined by the LLM 202 based on its knowledge and capabilities. In some instances, when responding to one or more user queries, LLM 202 can generate a dialogue that may include responses to the queries, request additional information from the user, refine the dialogue between the user and the Petrophysical Assistant 200, and / or relay and / or request any other relevant information. For example, if LLM 202 determines that a user query is too broad or vague to provide a meaningful response, LLM 202 can request additional information from the user to refine the query and generate a more meaningful response. As another example, if the Control LLM 202 determines that the query response(s) from a model(s) among the Scope-Specific LLMs 210 are too general (for example, lacking a minimum level of specificity and / or relevance) and / or too long, the Control LLM 202 can attempt to reduce or improve the query response(s) or request additional information from the user to improve the query and use the improved query to provide a more specific and / or shorter response.In some cases, LLM 202 may include one or more parameters that define any constraints it can use to determine whether a query and / or query response should be optimized. For example, if a query is too broad, the query response obtained by LLM 202 may be too general and / or too long. To determine whether a query is too broad and / or the query response is too general and / or too long, LLM 202 may include parameters that define a limit for the size of the response, a limit for the size of the query, a minimum limit for the size of the query, privacy for the response, privacy for the query, and / or any other parameters it can use to determine whether a query and / or query response needs to be expanded, reduced, and / or optimized. As mentioned previously, the LLM 202 can include a multi-pattern model 206, such as an image-to-text generation model, a video-to-text generation model, a speech-to-text generation model, and / or any other multi-pattern generation model. A multi-pattern model 206 can allow the Petrophysical Assistant 200 to convert input data from one pattern to another and / or output data from one pattern to another. For example, a multi-pattern model 206 can convert image data (e.g., a picture or video, etc.) provided by the user into text data that the Petrophysical Assistant 200 can use to understand a query and / or formulate a response. For example, as part of a query (and / or as part of information provided with a query), the user can provide image data (e.g., via the LLM 204 interface) to the Petrophysical Assistant 200, such as one or more graphs, maps, records, or petrophysical charts, and / or other graphical information. The multi-pattern model 206 can convert this image data into text data and provide the text data to the LLM controller 202.LLM 202 can use textual data to understand the query, formulate / reformulate the query, and / or provide a response to the query. In this way, complex visual data can be used effectively, ensuring that the nuances of context for each field are not lost or ignored. As another example, a user can submit (for instance, via the LLM Interface 204) a visual petrophysical record to the petrophysical assistant 200 along with a question about a portion of the visual petrophysical record. The multi-pattern form 206 can scan the visual petrophysical record and convert the visual petrophysical record (or the relevant portion of the visual petrophysical record) into text (for instance, by image-to-text conversion), providing information (for instance, description, translation, summary, details, etc.) about the visual petrophysical record (or the relevant portion of the visual petrophysical record), such as visual petrophysical record (or relevant portion of the visual petrophysical record) descriptors.The multi-pattern model 206 can provide the resulting text from the image data to the control LLM 202, which can be used to insert it with the query, formulate the query, complete the information contained in the query, provide additional context to the query, help identify the model(s) among the range-specific LLMs 210 to direct the query to, and / or provide query-related information that the model(s) among the range-specific LLMs 210 selected to answer the query can use to understand the query and / or provide a response to the query. As another example, the Multi-Mode Model 206 can convert audio received with an input query, such as speech or audio data, into text that describes the audio and / or provides information about the audio. The Multi-Mode Model 206 can provide the resulting audio text to the Control LLM 202, which can then use the resulting text and any other information in the query to understand the query, formulate / reformulate the query, complete the query, provide additional context, help identify the model(s) from among the range-specific LLMs 210 to direct the query to, and / or provide query-related information that the selected model(s) from among the range-specific LLMs 210 can use to answer the query. The Multi-Mode 206 model can be trained using petrophysical data of a specific format / type, such as the input mode. For example, to train the Multi-Mode 206 model to convert visual petrophysical data (e.g., images and / or videos such as logs, graphs, visual maps, etc.) into text describing that visual petrophysical data, the Multi-Mode 206 model can be trained using a dataset containing visual petrophysical data. In some cases, the training dataset (e.g., the visual petrophysical dataset) can include range-specific petrophysical data, general petrophysical data, field data, and / or any other petrophysical and / or field data. In some respects, textual information derived from the multi-pattern model 206 from other types of data (for example, image data, sound data, etc.) can be used to fine-tune the control LLM 202 and / or range-specific LLMs 210. By enabling models to learn from and adapt different types / patterns of data, the understanding of petrophysical ranges can be enhanced by the petrophysical assistant 200. In some examples, the LLM 202 controller, the LLM 204 interface, and / or the multi-pattern model 206 can be part of the same model. For example, the LLM 202 controller, the LLM 204 interface, and / or the multi-pattern model 206 can be implemented by the same model. As another example, the LLM 202 controller can include a base model, and the LLM 204 interface and / or the multi-pattern model 206 can include one or more headers or branches of the model. In other examples, the LLM 202 controller, the LLM 204 interface, and / or the multi-pattern model 206 can be part of different models. For example, the LLM 202 controller can be a different model from the model(s) that implement the LLM 204 interface and / or the multi-pattern model 206. In some examples, an administering LLM 202 and / or any of the scope-bound LLMs 210 can access Knowledge Base 230 information to use in understanding a query and / or formulating a query response. For example, when a model from among the scope-bound LLMs 210 receives a query, the model can perform a semantic search of the relevant data in Knowledge Base 230 to identify the best and / or most relevant information in Knowledge Base 230 to use in formulating a query response. Similarly, an administering LLM 202 can access Knowledge Base 230 to understand a query and / or obtain information that the administering LLM 202 can use, along with any query responses from any of the scope-bound LLMs 210, to formulate / rewrite a query response for the user and / or supplement, review, and / or combine one or more query responses from one or more scope-bound LLMs 210.In some cases, the meaning of the content (e.g., sentences, words, etc.) in Knowledge Base 230 can be encoded into coded or structured data (e.g., vectors that encode the meaning of the content such as sentences, etc.) which can be used to find information while conducting a semantic search in Knowledge Base 230. For example, when a model (e.g., a control LLM 202 and / or any of the range-bound LLMs 210) receives a query, the model can encode the query into a coded query (e.g., a vector, and / or structured data that encodes the query and / or the query meaning, etc.) and use it to search for the encoded data in Knowledge Base 230 to find a semantic search result. In some cases, the model can encode the query into a vector that encodes the semantic meaning of the query and use the encoded vector to identify one or more encoded vectors in Knowledge Base 230 that encode the semantic meaning of the content in Knowledge Base 230, such as sentences, which best match the encoded query.The one or more encoded vectors that best match the encoded query may include, but are not limited to, the encoded vector from Knowledge Base 230 that has been identified as the best / highest match for the encoded query, or the best number n of matching encoded vectors from Knowledge Base 230, or the encoded vector(s) from Knowledge Base 230 that have at least a serious match value with the encoded query, or any other number of encoded vectors. In some cases, a model (for example, the Control LLM 202 and / or any of the Scope-Specific LLMs 210) can use Knowledge Base 230 to search and the model's acquired abilities (for example, acquired knowledge, acquired reasoning abilities, acquired semantic comprehension abilities, acquired pattern recognition abilities, etc.) to assimilate the information in Knowledge Base 230, analyze it, understand it, interpret it, verify it, organize it, filter it, sort it, and / or identify it. Knowledge Base 230 can include any petrophysical information such as, but not limited to, field-specific knowledge, instrument-specific knowledge, context-specific knowledge, general and / or scope-specific petrophysical knowledge, and / or any other petrophysical information.In some examples, a knowledge base can include 230 information repositories (e.g., database(s), data store(s), library, dataset, collection of documents, etc.) such as, but not limited to, one or more journal articles, scientific papers, field reports, question-and-answer pairs, petrophysical and / or wellbore records, instrument data, field data, training courses / materials, tutorials, books, manuals, instrument data, data collection and processing, drilling reports, visual data (e.g., graphs, charts, maps, pictures, videos, records, etc.), publications, textbooks, internal knowledge materials, historical data, statistics, database knowledge, comments, records, research data, user letters and / or feedback, and / or any other petrophysical data and / or knowledge materials. In some cases, Knowledge Base 230 may include similar results and / or inferences from one or more petrophysical situations (e.g., similar, different, relevant, etc.) from petrophysical cases, questions, contexts, conditions, or events encountered in the field. With the information from Knowledge Base 230, LLM Control 202 can improve its responses to petrophysical queries, thereby enhancing accuracy and reliability. Knowledge Base 230 can also be updated as needed to add or remove specific information. For example, it can be updated with additional materials, results, inferences, field information, and / or relevant information as required or desired. As described above, the Petrophysics Assistant 200 can possess and / or offer general knowledge, petrophysics-specific knowledge, tool-specific knowledge, field-specific knowledge, and / or any other relevant knowledge. Furthermore, the Petrophysics Assistant 200 can include natural language processing capabilities, multi-mode processing capabilities (e.g., visual data processing capabilities, audio data processing capabilities, etc.), reasoning capabilities, learning capabilities, response / data synthesis and / or formulation capabilities (e.g., cognitive response synthesis capabilities, etc.), and / or any other capabilities. The Petrophysics Assistant 200 can provide feedback on the working condition, answer specific questions, offer advice and / or suggestions to users, make decisions, identify problems, provide information to help one or more users manage the wellbore environment and / or wellbore operations, and / or provide any other assistance to users with respect to petrophysics, well / field drilling, associated tools, and / or any other relevant knowledge. For example, a user, such as a scientist or engineer working at a well site, could use the Petrophysical Assistant 200 as an AI tool to obtain feedback on a task or function, ask questions about the well site (e.g., associated conditions, formation characteristics, events, petrophysical issues, measurements, petrophysical inferences / results, analysis, etc.), ask questions about a function or task at the well site, ask questions related to a specific context (e.g., field context, instrument context, use case, situation, etc.), make decisions, identify problems, obtain relevant information, conduct consultations, receive suggestions, and so on. To clarify, if a user has a question about a measurement, inference, petrophysical condition, or something similar, they can submit the question to the Petrophysical Assistant 200.The Petrophysical Assistant's Control LLM 202 can identify a subset of range-specific LLMs from among the 210 range-specific LLMs to direct the question to, and send a call, command, instruction, and / or message to the identified subset of range-specific LLMs, which then triggers / executes the subset of range-specific LLMs to generate one or more relevant responses to the question. A subset of range-specific LLMs can then provide one or more relevant responses to Control LLM 202, which can use one or more of these responses to formulate a response to the user's query. The LLM 202 can then present the formulated response to the user via the LLM 204 interface. The LLM 202 can create a dialogue for the user to present the formulated response, ask follow-up questions, request additional information, receive additional or follow-up questions from the user, provide further assistance to the user, etc., as described here. The Petrophysical Assistant 200 can run on any computing device, such as a server, a computing instrument, a client-architecture device, and / or any other computing and / or edge device. For example, the Petrophysical Assistant 200 can be compressed to run on edge devices with fewer computing resources (e.g., greater computing resource constraints) compared to high-end devices such as servers. Furthermore, the Control LLM 202 and / or any of the Scope-Specific LLMs 210 can be trained using data generated and / or collected by a human and / or data generated and / or collected by a model. For example, in some cases, the quirky LLM 202 and / or any of the range-bound LLMs 210 can be trained using question-and-answer pairs created and / or compiled by a human and / or a model. In some examples, the LLM can be used to generate question-and-answer pairs that can then be used to train the quirky LLM 202 and / or any of the range-bound LLMs 210 (for example, with or without any question-and-answer pairs created / compiled by a human). Figure 3 is a diagram illustrating a 300 demonstration system for generating training data used to train the 202 control LLM and / or any of the 210 range-bound LLMs, based on some current detection examples. In this example, LLM 304 can be used to generate training data 306, which is then used to train the 202 control LLM and / or any of the 210 range-bound LLMs. LLM 304 can generate training data 306 based on input data 302, which includes the relevant knowledge / information. In some cases, 302 input data may include, but are not limited to, one or more journal articles, scientific papers, field reports, question-and-answer pairs, petrophysical and / or well logs, instrument data, field data, training courses / materials, tutorials, books, manuals, instrument data, data collection and processing, drilling reports, publications, textbooks, internal knowledge materials, historical data, statistics, database knowledge, comments, recordings, research data, user letters and / or notes, and / or any other petrophysical data and / or knowledge materials. Furthermore, 306 training data may include any relevant structured knowledge / information. For example, in some cases, 306 training data may include question-and-answer pairs generated from 302 input data. In some examples, the data used to generate Input Data 302 can be categorized into subject areas, topics, scenarios, or related use cases, and / or any other category. For example, to generate Input Data 302, a petrophysics dataset (e.g., papers, books, brochures, documents, etc.) and / or parts thereof (e.g., by LLM 304 or another model) can be categorized into subject areas. Each item can be categorized within one or more of the relevant subject areas. For example, a petrophysics research paper can be categorized within a single subject area, or, in some cases where the petrophysics research paper relates to multiple subject areas, within multiple subject areas. Thus, Input Data 302 can include data categorized into different subject area catchments. LLM 304 can process Input Data 302 to generate Training Data 306.For example, LLM 304 can process data in each taxonomic catchment to generate question-and-answer pairs related to that taxonomic catchment (e.g., related to the subject area associated with that taxonomic catchment). In this example, the question-and-answer pairs in the training data 306 can then be used to train the control LLM 202 and / or any of the range-bound LLMs 210. For example, question-and-answer pairs from a taxonomic catchment related to a specific range can be used to train a range-bound LLM associated with that specific range (and in some cases, the control LLM 202). For example, question-and-answer pairs from a taxonomic catchment related to resistivity can be used to train the resistivity model 212, while question-and-answer pairs from a taxonomic catchment related to NMR can be used to train the NMR model 214. In some cases, the LLM 202 can be trained using general and / or scope-specific question-and-answer pairs found in the 306 training data. For example, the LLM 202 can be trained using question-and-answer pairs from a classification pool related to general petrophysical knowledge. In other cases, the LLM 202 can be further or alternatively trained using question-and-answer pairs from one or more classification pools related to one or more subject areas. Figure 4 is a diagram showing an illustrative use of the Petrophysical Aid 200, based on some examples of current disclosure. As shown, a 402 user (e.g., a petrologist, engineer, technician, agent, etc.) can access the Petrophysical Aid 200 from a 404 computer. A 404 computer can include any computer capable of running a model such as, but not limited to, a server, an edge device (e.g., a laptop, desktop computer, tablet, or a tool with computing capabilities, etc.), or any other computing device. User 402 can use computer 404 to send a 406 query to Petrophysics Assistant 200, which Petrophysics Assistant 200 can then use to generate a 408 response for User 402. In some cases, when generating the 408 response, Petrophysics Assistant 200 can perform one or more searches in Knowledge Base 230, as previously described. Furthermore, in some instances, Petrophysics Assistant 200 can initiate a dialogue with User 402. For example, Petrophysics Assistant 200 can receive the 406 query, provide the 408 response, receive a follow-up query (and / or request additional information), provide a follow-up response, and so on. To illustrate this, suppose query 406 includes the question, "Why is neutron porosity different from NMR porosity?" Petrophysics Assistant 200 could receive query 406 and generate a response 408, such as, "Neutron porosity is affected by lithological properties and specific fluid properties. NMR porosity is affected by lithological properties, stratification, and recording speed. I observe that for low-gamma-ray conductive fluids, which are Braine in a silicate-crystalline configuration, neutron porosity is much more pronounced than NMR porosity. This suggests that recording speed might be the issue for NMR porosity. Why not try slowing down the NMR data recording speed?" Petrophysics Assistant 200 could then provide such a 408 response to user 402. In some cases, user 402 might want to send a follow-up query such as, for example, "Well, re-recording at a slower speed matches much better, but the transmittance seems high for this field. What do you think?" Similarly, the Petrophysical Assistant 200 can generate a response to a follow-up query, and in this way engages in a dialogue with the 402 user via the 404 computer. Figure 5 illustrates an example of a 510 neural network, based on some of the current detection examples. The 510 neural network can be used to implement any of the models described here, such as the 202 control LLM, the 206 multi-mode model, any of the 210 range-bound LLMs, the 304 LLM, and so on. As shown in this example, the 510 neural network includes an input layer, 502, to process the input data. The 510 neural network also includes hidden layers, 504a through 504n (collectively referred to hereafter as "504"). The 504 hidden layers can include n hidden layers, where n is an integer greater than or equal to one. The number of hidden layers can include any number of layers required to achieve a desired processing result and / or a specific display purpose. The 510 neural network includes an output layer, 506, which provides the output resulting from the processing performed by the 504 hidden layers. The 510 neural network in this example is a multi-layered neural network of interconnected nodes. Each node can point to a piece of information. The information represented by the nodes is shared between the different layers, and each layer retains the information during processing. In some cases, the 510 neural network can be a feedforward network, in which case there are no feedback loops, and the network outputs are returned to the same source. In other cases, the 510 neural network can be a recurrent neural network, which can contain loops that allow information to be passed through the nodes as it is read from the input. Information can be exchanged between nodes via node-to-node interconnects across multiple layers. Input layer 502 nodes can activate a group of nodes in the first hidden layer, 504a. For example, as shown, each input layer 502 node is connected to each hidden layer 504a node. The hidden layer 504a nodes can transform the information from each input node by applying activation functions to the information. The information derived from this transformation can be passed to the next hidden layer node (e.g., 504b) and activated, allowing the nodes to then execute their assigned functions. These functions include convolution, expansion, data transformation, aggregation, and / or other appropriate functions. The output of the hidden layer (e.g., 504b) can then activate the next hidden layer node (e.g., 504n), and so on. The output of the last hidden layer can activate one or more output layer 506 nodes, and at this stage the output is provided.In some cases, while the nodes (for example, nodes 508a, 508b, 508c) in the 510 neural network are shown as having many output lines, the node actually has a single output, and the lines shown in the output images of the node represent the same output value. In some cases, each node or internode can have a weight representing a set of variables derived from the 510 neural network's training. For example, an internode can represent information about the interconnected nodes. This internode can also have a numerical weight that can be adjusted (for example, based on a training dataset), allowing the 510 neural network to adapt to inputs and enabling it to learn as more information is processed. The 510 neural network can be pre-trained to process features taken from the data in the input layer 502 using the various hidden layers 504 in order to provide the output via the output layer 506. In one example where the 510 neural network is used to output textual answers, the 510 neural network can be trained using training data that includes pairs of questions and answers. In some cases, the 510 neural network can adjust node weights using a training process called backpropagation. Backpropagation can involve forward scrolling, a loss function, backswiping, and weight updates. These forward scrolling, loss function, backswiping, and variable updates are performed in a single training iteration. The process can be repeated a certain number of times for each set of training media data until the layer weights are precisely adjusted. For example, forward passing could involve passing training data through a 510 neural network. The weights could be initially randomly assigned before the 510 neural network is trained. For the first training iteration of the 510 neural network, the output could include values that do not favor any particular class because the weights were initially randomly assigned. For example, if the output is a vector with probabilities of having different outputs, the probability value for each of the different outputs could be equal or at least very similar (for example, for ten possible outputs, each output could have a probability value of 0.1). With the initial weights, the 510 neural network might fail to identify low-level features and therefore might not identify a specific class. A loss function could be used to analyze the errors in the output. Any suitable definition of a loss function can be used. Loss (or error) can be high for the initial training set (e.g., images) as long as the actual values will differ from the predicted output. The goal of training is to minimize the loss so that the predicted output matches a target or ideal output. The 510 neural network can perform a backhaul by identifying the inputs (weights) that contribute most to the network's loss, and then adjusting the weights so that the loss decreases and eventually becomes minimal. A derivative of the loss for each weight can be calculated to identify the weights that contributed the most to the loss in the 510 neural network. After calculating the derivative, the weights can be updated by updating the filter weights. For example, the weights can be updated so that they change in the opposite direction of the gradient. The learning rate can be set to any appropriate value, where a high learning rate involves larger weight updates and a low value indicates smaller weight updates. The 510 neural network can encompass any suitable neural network or deep learning method. One example is a convolutional neural network (CNN), which has an input layer and an output layer, with several hidden layers in between. These hidden layers in a CNN include a series of convolutional, nonlinear, clustering (for reduction), and full-correlation layers. In other examples, the 510 neural network can represent any other neural network or deep learning method, such as a switching network, an auto-encoder, deep belief networks (DBNs), a recurrent neural network (RNN), a length-limiting learning network (LLM), and so on. Figure 6 is a diagram illustrating a 600 model structure that can be used to implement an LLM, such as a 202 control LLM, a 210 range-bound LLM, a 304 LLM, etc. In this example, the 600 model structure represents a switching network structure that can be used to implement an LLM. As shown, the 600 model structure can include 602 input embeddings used as inputs to the model. These 602 input embeddings provide values that represent words and / or phrases, such as numbers or vectors that represent words and / or phrases. 602 input embeddings can act as a dictionary, helping the model understand word meanings by placing them in an embedding space where similar words are located near each other. In some cases, the model can learn to create 602 input embeddings during training so that similar vectors represent words with similar meanings. The model can use 604 positional coding to encode the position of each word in the input sequence from the 602 input embeddings as values such as a set of numbers or a vector, etc. The values generated by 604 positional coding can be fed into the model along with the 602 input embeddings. By incorporating 604 positional coding into the 600 model structure, the model can more effectively understand word order in a sentence and generate grammatically correct and semantically meaningful outputs. The model architecture can include a 600-606 encoder(s) used to process localized 602-coded input embeddings and generate 608-coded embeddings. The 606 encoder(s) can be part of the model that processes the input text and generates hidden states that capture the text's meaning and context. For example, the 606 encoder(s) might include a feedforward neural network that forms part of the transformer model. In some instances, the 606 encoder(s) can implement multiple encoding layers. In some cases, the 606 encoder(s) might first decode the input text into a series of symbols, such as individual words or individual subwords. The 606 encoder(s) can then apply one or more self-attention layers, which can generate hidden states representing the input text at different levels of abstraction. In this way, the 606 encoder(s) can create 608 embeddings (for example, a vector, a set of values, etc.) that represent the semantics and positions of words in one or more sentences. The model structure can include 600 output embeddings (612), which can contain values representing words and / or phrases, such as numbers or vectors. These 612 output embeddings can be similar to the 602 input embeddings and can also be processed using positional coding (614) to encode the position of each word in a sequence of 612 output embeddings as values such as a set of numbers or a vector, etc. This helps the model understand the word order in a sentence. 612 output embeddings can be used during the model training phase, but they can also be used during the extrapolation phase. During training, a loss function can be calculated based on the 612 output embeddings and used to update the model parameters to improve model accuracy. During the extrapolation phase, 612 output embeddings can be used to generate the output text by associating the model's expected probabilities for each symbol with the corresponding symbol in the vocabulary. Locally encoded input embeddings (e.g., embeddings 602) and locally encoded output embeddings (e.g., embeddings 612) can be fed into a decoder unit(s) 610, which is used to generate an output sequence based on the encoded input sequence. During training, the decoder unit(s) 610 can learn to guess the next word in the sequence by considering the words that precede it. In some examples, the decoder unit(s) 610 can generate natural language text based on the input sequence and any learned context. A 610 decoder(s) can generate 616 embeddings and feed these embeddings into one or more 618 network layers. In some examples, one or more 618 network layers might include a linear layer and a softmax function. The linear layer can map the 616 embeddings generated by the 610 decoder(s) to a higher-dimensional space, which can then transform the 616 embeddings into their original input space. The softmax function can then be applied to generate a probability distribution for each output symbol in the vocabulary, potentially resulting in a 620 output. In some examples, the 620 output might contain output symbols with probabilities. Figure 7 is a flowchart showing an illustrative process 700 for implementing a petrophysical aid (e.g., petrophysical aid 200), according to some current detection examples. In framework 702, process 700 may involve receiving, via an administer language model (e.g., administer LLM 202) configured to perform natural language processing, a query relating to one or more subject areas. In some examples, one or more subject areas may include one or more petrophysical subject areas such as, but not limited to, resistivity, NMR, porosity, continuity, fluid contact surfaces, permeability, stone properties, density, saturation (e.g., water / fluid saturation, hydrocarbon saturation, etc.), pressure, magnetism and / or electromagnetism (EM), and / or composition testing, among others. In other examples, one or more subject areas may include any other subject areas within petrophysics and / or in any other field. In framework 704, process 700 can involve selecting, based on one or more subject areas related to the query and the relevant scope-specific knowledge for each scope-specific language model from among a set of scope-specific language models (e.g., scope-specific LLMs 210), one or more scope-specific language models from among a set of scope-specific language models to answer the query.In some examples, selecting one or more domain-specific language models to answer a query might involve the arbitrary language model identifying which subset of the domain-specific language models can or should answer the query related to one or more subject areas; have expertise / knowledge in one or more subject areas; have been trained to answer queries in one or more subject areas; have been trained with information related to one or more subject areas; have reasoning and / or cognitive abilities in one or more subject areas; and / or are best suited to answer the query related to one or more subject areas. The arbitrary language model then selects the subset of domain-specific language models to answer the query. The arbitrary language model can then direct the query to the subset of domain-specific language models to run the query (or attempt to answer) that subset. In some examples, each scope-specific language model (LSM) within a set of LLMs can include a neural network, such as an LLM, trained with information about a different petrophysical domain and / or configured to answer queries related to that different petrophysical domain. For example, each LLM can be trained with information about a different petrophysical domain(s) and / or configured to answer queries related to that different petrophysical domain(s). In some cases, each neural network (e.g., each LLM) can be trained with different context-specific information and / or configured to answer queries related to that different context-specific information. This different context-specific information can include, but is not limited to, tool-specific information, wellbore-specific information, use-state-specific information, and so on. In some respects, an administer language model can include a neural network (e.g., an LLM) trained with petrophysical information, and each domain-specific language model within a set of domain-specific language models can be trained with more specific petrophysical information than the administer language model. For example, an administer language model can be trained with general petrophysical information, such as petrophysical information from one or more petrophysical domains (e.g., topics, subject areas, concepts, principles, use cases, etc.) that is more general than (e.g., less specific than, less detailed than, less comprehensive than, less advanced than, less in-depth than, at a higher level than, more general than, etc.) the more specific petrophysical information used to train each of the domain-specific language models. In other words, the petrophysical information used to train a control language model can be more general than the petrophysical information used to train each specific domain language model, at least with respect to the domain of petrophysics in which the domain language model is trained / customized. In some aspects, an administer language model and / or any of the range-specific language models may include a network of adapters. In some examples, a set of scope-specific language models may include a resistivity model (e.g., Resistivity Model 212) set up to answer questions about resistivity, an NMR model (e.g., NMR Model 214) set up to answer questions about NMR, a porosity model (e.g., Porosity Model 216) set up to answer questions about porosity, a fluid contact surfaces model (e.g., Fluid Contact Surfaces Model 220) set up to answer questions about fluid contact surfaces, a permeability model (e.g., Permeability Model 222) set up to answer questions about permeability, a configuration test model (e.g., Configuration Test Model 224) set up to answer questions about configuration testing, and / or a continuity model (e.g., Continuity Model 218) set up to answer questions about continuity. In framework 706, the process can involve sending, to one or more scalable language models, a request to answer the query. For example, a scalable language model can send a call, message, command, and / or instruction to one or more scalable language models configured to run one or more scalable language models on a machine that hosts a set of scalable language models and generate (or attempt to generate) one or more of the relevant responses to the query. In framework 708, process 700 can involve the creation, by an administering language model, of a query response based on one or more query responses received from one or more scope-defined language models. For example, an administering language model can receive one or more query responses from one or more scope-defined language models, and use one or more query responses to formulate a response by combining one or more query responses, summarizing one or more query responses, revising one or more query responses, preparing information from one or more query responses into a dialog-style response, and / or preparing a query response in another way. In some examples, an optimization language model can generate a query response based on information from one or more responses and petrophysical information that the model has learned. For example, the optimization language model can use information from each of one or more responses (or at least some of them) to generate a response that is partly based on petrophysical information learned and / or identified by the model. In some respects, the process can involve 700 relevant search operations, by each of one or more domain-specific language models and based on the query, in a knowledge base (e.g., Knowledge Base 230) containing petrophysical information, and the creation, by one or more domain-specific language models, of one or more responses based on each relevant search in the knowledge base containing petrophysical information. In some respects, one or more scoped language models may include a subset of scoped language models from within a larger set of scoped language models, and one or more responses may include multiple responses from different scoped language models within that subset of scoped language models. In some examples, generating a response to a query may involve combining information from multiple responses into a single, unified response. In some respects, the logic language model may include a model (e.g., the multi-pattern model 206) configured to convert image data into textual data describing the image data, and the query may include visual information (e.g., image data) collected from a wellbore site. In some cases, the visual information may include, but is not limited to, a petrophysical visual record, a petrophysical graph, a petrophysical map, a petrophysical diagram, and / or other graphical information. For example, the visual information may include a visual record obtained from logging tools 126. In some cases, the visual record may include a reference to a portion of the visual record relevant to the query and / or specific to consider in relation to the query, such as a specific portion of the visual record, a comment in the visual record, a marker in the visual record, metadata added to the visual record, a note in the visual record, the formatting of a portion of the visual record, the color coding of a portion of the visual record, and so on. In some examples, the selection of one or more scope-specific language models and / or the generation of the response can be further dependent on visual information. For example, a model configured to convert image data to text can convert visual information into text that describes, summarizes, identifies, and / or categorizes the visual information (and / or a portion thereof). The converted text can then be included in the query and / or submitted with the query to one or more scope-specific language models, which can use the converted text and any other information in the query to generate the relevant response to the query. In some examples, an administer language model and a set of scalable language models can be part of a petrophysical assistant (e.g., the petrophysical assistant 200) and can run on a computing device, such as a server, or an instrument (e.g., a recording device, a sensor system, etc.) with computing capabilities, a client-architecture device, an edge device, and / or any other computing device. For example, the petrophysical assistant can be compressed to run on edge devices with fewer computing resources (e.g., greater computing resource constraints) compared to high-end devices such as servers. In some cases, an edge device may not have sufficient computing resources to run both an administer language model and a set of scalable language models simultaneously.However, by allowing the logic controller to selectively route queries to a subset of scalable LAN models, and thus run only a subset of scalable LAN models on the edge machine at a given time (for example, for a specific query / call), the logic controller can reduce the amount of resources used on the edge machine to answer a query at any given time. This, in turn, allows the edge machine to run the selected subset of scalable LAN models at a given time despite the edge machine's resource constraints. This prevents errors / crashes that would occur if the edge machine attempted to run every scalable LAN model set simultaneously (or a model with the size / parameters and / or capabilities of every scalable LAN model set) given the edge machine's resource limitations. It also reduces response time when running the petrophysical assistant (compared to running every scalable LAN model set simultaneously or running a model with the size / parameters and / or capabilities of every scalable LAN model set), among other benefits. Figure 8 illustrates the architecture of a schematic 800 computer that can be used to implement any of the systems and techniques described here. In some examples, the computer architecture can be integrated into the electromagnetic imaging tools described here. Furthermore, the computer can be configured to implement machine learning techniques for controlling the composition of borehole images, as described in this document. The components of the 800 computing device architecture are shown electrically connected to each other using an 805 interface, such as a bus. The 800 illustrative computing device architecture includes an 810 CPU (or processor) and an 805 interface that connects the various computing device components, including the 815 computing device memory (such as 820 read-only memory (ROM) and 825 random-access memory (RAM), to the 810 processor. The 800 PC architecture can include a high-speed cache memory directly connected to, located near, or integrated with the 810 processor. The 800 PC architecture can copy data from the 815 memory and / or the 830 storage device into the 812 cache for faster access by the 810 processor. This can enhance performance and prevent the 810 processor from lagting while waiting for data. These and other modules can be configured to control or control the 810 processor to perform various actions. Other 815 PC memory modules may also be available. The 815 memory can include several different types of memory with varying performance characteristics. The 810 processor can include any general-purpose processor and computer components or software services, such as Service 1 832, Service 2 834, and Service 3 836 stored in an 830 storage device, configured to control the 810 processor, as well as a special-purpose processor where the program instructions are included in the processor design.The 810 processor can be a standalone system, containing multiple cores or processors, a bus, a memory controller, a buffer, and so on. A multi-core processor can be symmetric or asymmetric. To facilitate user interaction with the 800 PC architecture, the 845 input device can represent any number of input mechanisms, such as a microphone for speech input, a touchscreen for motion or graphics input, a keyboard, a mouse, motion input, speech input, and so on. The 835 output device can also be one or more of several output mechanisms familiar to those skilled in the field, such as a monitor, a projector, a television, an amplifier, and so forth. In some cases, multi-mode PCs can allow the user to provide multiple types of inputs to connect to the 800 PC architecture. The 840 Interface can generally control and manage user inputs and PC outputs. There is no restriction to any specific hardware configuration, and therefore the basic features described here can easily be replaced by improved hardware or firmware configurations during development. The 830 storage device is a non-volatile memory device that can be a hard disk or other types of computer-readable media that can store data accessible by a computer, such as magnetic cassettes, flash memory cards, solid-state drives, digital multipurpose discs, cartridges, random access memories (RAMs) 825, read-only memory (ROM) 820, and hybrids thereof. The 830 storage device can include services 832, 834, and 836 to control the 810 processor. Hardware component modules or software modules are described. The 830 storage device can be connected to the 805 computer interface. In one aspect, a hardware component module that performs a specific function may include software components stored on computer-readable media related to the necessary hardware components, such as the 810 processor, the 805 interface, the 835 output device, and so on, to perform the function. To clarify the explanation, in some cases the current technology can be presented as including independent functional templates that include functional templates that include devices, hardware components, steps or routines in a method implemented in a program, or combinations of hardware components and programs. In some cases, storage devices, media, and computer-readable memory devices may include a cable or wireless signal containing bits of current and the like. However, when referring to them, non-temporary computer-readable storage media clearly include media such as power, carrier signals, electromagnetic waves, and the signals themselves. The methods described in the preceding examples can be implemented using computer-executable instructions stored or made available from computer-readable media. These instructions include, for example, instructions and data that drive or configure a general-purpose computer, a special-purpose computer, or a processing device to perform a specific function or set of functions. The computer resources used are accessible through a network. Computer-executable instructions may be, for example, binary numbers, or instructions in an intermediate format such as assembly language, built-in programs, source code, etc. Examples of computer-readable media that can be used to store instructions, and / or information used, and / or information generated during the methods described include magnetic or optical motors, flash memory, USB devices with hard drives, network-attached storage devices, etc. Devices that implement the methods described in the aforementioned detection techniques include hardware components, built-in software, and / or software components, and can take a wide variety of forms. Examples of these forms include laptops, smartphones, mini PCs, personal digital assistants (PDAs), rack-mounted devices, standalone devices, and so on. The functionalities described here can also be implemented in peripheral connectors or add-on cards. Furthermore, these functionalities can be implemented on a circuit board within different chips or through different processes within a single device, for example. Instructions, the means of transmitting these instructions, the computing resources for their execution, and other structures to support said computing resources are illustrative means of providing the functions described in the disclosure. In the preceding description, aspects of the requirement are described by reference to specific examples and aspects. However, those skilled in the field will recognize that the requirement is not limited to these. Therefore, while examples and illustrative aspects of the requirement are described in detail here, it will be understood that the principles revealed may be incorporated and used differently, and that the accompanying safeguards are to be considered to encompass these variations, except as otherwise specified in the preceding text. The various features and aspects of the subject described above may be used individually or in combination. Furthermore, the examples and aspects of the systems and techniques described here may be used in any number of environments and applications other than those described herein without straying from the broader scope and content of the description. Therefore, the specification and drawings should be viewed as illustrative, not exhaustive. For the sake of clarity, the methods are described in a specific order. It should be understood that in alternative examples, the methods may be implemented in a different order than that described. When components are described as being "configured" to perform certain operations, this configuration can be implemented, for example, by designing electronic circuits or another computer component to perform the operation, by programming programmable electronic circuits (for example, small-scale data processors, or other suitable electronic circuits) to perform the operation, or any combination thereof. Many of the logic blocks, modules, circuits, and various algorithmic steps described here, with reference to the examples provided, can be implemented as electronic computer components, computer software, firmware, or combinations thereof. To clearly illustrate the interchangeability between computer components and software, many of the components, blocks, modules, circuits, and illustrative steps above are described in terms of their function. Whether that function is implemented as computer components or software depends on the specific usage and design constraints imposed on the entire system. Experienced practitioners may implement the described function in several ways for each specific use, but the implementation decisions mentioned should not be interpreted as a departure from the current application. The techniques described here can be implemented in electronic computer components, computer software, firmware, or any combination thereof. These techniques can be implemented in a variety of ways, such as in multifunction computers, walkie-talkie headsets, or multifunctional integrated circuit devices, including their use in walkie-talkie headsets and other devices. Any features described as modular or component-based can be implemented together in an integrated logic device or separately as separate but interoperable logic devices. If implemented in software, the techniques can be realized, at least in part, by a computer-readable data storage medium containing program code that includes instructions which, when executed, perform one or more of the methods, algorithms, and / or operations described above. The computer-readable data storage medium can be part of a computer software product, which may include packaging materials. A computer-readable medium can include memory or data storage media, such as random access memory (RAM), synchronous dynamic random access memory (SDRAM), read-only memory (ROM), non-volatile random access memory (NVRAM), electrically erasable programmable read-only memory (EEPROM), flash memory, magnetic or optical data storage media, and the like. Alternatively, these technologies can be achieved, at least in part, by a computer-readable communication medium that carries or transmits program code in the form of instructions or data structures that can be accessed, read, and / or executed by a computer, such as propagating signals or waves. Detection methods and devices can be implemented in networked computing environments with many types of computer system architectures, including personal computers, handheld devices, multiprocessor systems, programmable or microprocessor-based consumer electronics, network computers, microcomputers, mainframe computers, and the like. These methods can also be implemented in distributed computing environments where tasks are performed by local and remotely connected processing devices that are linked (either by wired connections, wireless connections, or a combination thereof) via a communications network. In a distributed computing environment, software modules can be placed on both local and remotely connected memory storage devices. In the following description, terms such as "upper," "upper," "below," "down," "above," "beneath," "bottom of the well," "top of the well," "longitudinal," "lateral," and the like, as used here, should mean relative to the bottom and the farthest extent of the surrounding wellbore, even though the wellbore or parts thereof may be skewed or horizontal. Similarly, transverse, axial, lateral, longitudinal, radial, and so on, mean directions relative to the direction of the wellbore or tool. The term "connected" is defined as being connected, either directly or indirectly through overlapping components, and need not be limited to physical connections. A connection may make objects permanently connected or detachable. The term "outside" refers to an area beyond the outer boundaries of a physical object. The term "inside" indicates that at least part of the area is partially included within a boundary formed by the object. The term "largely" is defined as essentially corresponding to a given dimension and shape, or another word that modifies it significantly, so that the component does not need to be precise. For example, "largely cylindrical" means that an object resembles a cylinder but incorporates one or more deviations from a true cylinder. The term "radius" means that the object is largely along the radius of its center, or has a directional component along the radius of its center, even if the object is not exactly circular or cylindrical. The term "axial" means that the object is largely along the axis of its center. In the absence of a specific axis, "axial" refers to the longer axis of the object. Although a variety of information is used to explain aspects within the scope of the Attached Protection Elements, no restrictions should be placed on the elements based on specific attributes or arrangements, as a person with average technical skills can deduce a wide range of applications. Furthermore, although part of the subject matter is described in terms of structural attributes and / or method steps, it should be noted that the subject matter of the Attached Protection Elements is not necessarily limited to those attributes or steps described. These functional tasks can be distributed differently or implemented in components other than those specified here. The attributes and steps are revealed as potential components of systems and methods that fall within the scope of the Attached Protection Elements. The language of a safeguard or other disclosure language that refers to "at least one of" a set and / or "one or more" of a set indicates that one or more members of the set (in any combination) satisfy the safeguard. For example, safeguard language referring to "at least one of A and B" or "at least one of A or B" means A, B, or A and B. In another example, safeguard language referring to "at least one of A, B, and C" or "at least one of A, B, or C" means A, B, C, A and B, A and C, B and C, or A, B, and C. Furthermore, the language "at least one of" a set and / or "one or more" of a set does not restrict the set to the elements included within it. For example, safeguard language referring to "at least one of A and B" or "at least one of A or B" could mean A, B, or A and B, and could also include elements not included in set A and B. The illustrative aspects of the disclosure include: Aspect 1. A method that includes: receiving, by an arbitrary language model configured to perform natural language processing, a query relating to one or more subject domains; based on the one or more subject domains related to the query and the relevant scope-specific knowledge for each scope-specific model from among a set of scope-specific models, selecting one or more scope-specific models from the set of scope-specific models to answer the query; sending, to one or more scope-specific models, a request to answer the query; and creating, by the arbitrary language model, a response to the query based on one or more responses to the query received from one or more scope-specific models. Side 2. The method according to Side 1, whereby each range-bound language model from the set of range-bound language models includes at least one neural network that has been trained with information about a different petrophysical domain and configured to answer queries related to the different petrophysical domain. Side 3. The method according to either Side 1 or 2, where each range-specific language model from the range-specific language model set includes at least one neural network that has been trained with different context-specific information and configured to answer queries associated with the different context-specific information, where the different context-specific information includes at least one tool-specific information and one well-hole-specific information. Aspect 4. The method according to any of the aspects from 1 to 3, where the computational language model can be trained using petrophysical information, and each range-bound language model from among the set of range-bound language models can be trained using more specific petrophysical information compared to the computational language model. Aspect 5: The method according to any of the aspects from 1 to 4, where the query response can be created based on the information received from one or more responses and the petrophysical information that the computational language model has learned to create the response. Aspect 6. The method according to any of the aspects 1 to 5, where one or more of the scope-bound language models include a subset of scope-bound language models from among the scope-bound language models set, and one or more of the responses include multiple responses from different scope-bound language models from the subset of scope-bound language models, and where the creation of the query response involves combining information from the multiple responses into a combined response. Aspect 7. The method according to any of the aspects from 1 to 6, where the control language model can additionally include a model configured to convert image data into textual data describing the image data, where the query includes visual information collected from the location of a well hole, and where at least one of the selection of one or more scoped language models and the creation of the response are additionally based on the visual information. Side 8. The method according to Side 7, where the visual information includes at least one petrophysical record, petrophysical map, petrophysical diagram, and petrophysical photograph. Aspect 9. The method according to any of the aspects from 1 to 8, where at least one of the control language model and one or more of the range-specific language model sets include a network of converters, and where the range-specific language model set includes at least one resistivity model set up to answer resistivity-related questions, a nuclear magnetic resonance (NMR) model set up to answer NMR-related questions, a porosity model set up to answer porosity-related questions, a fluid contact surface model set up to answer fluid contact surface-related questions, a permeability model set up to answer permeability-related questions, a configuration test model set up to answer configuration test-related questions, and a continuity model set up to answer continuity-related questions. Aspect 10. The method according to any of the aspects 1 to 9 also includes: conducting a relevant search, based on the query, by each of one or more scalable language models, in a knowledge base containing petrophysical information; and generating, by one or more scalable language models, one or more responses based on each relevant search in the knowledge base containing petrophysical information. Aspect 11. A system that includes: memory; and one or more memory-associated processors, one or more of which processors are configured to: receive, by means of a control language model configured to perform natural language processing, a query relating to one or more subject domains; and, based on the one or more subject domains associated with the query and the relevant scope determinant knowledge for each scope determinant model from among a set of scope determinants, select one or more scope determinants from among the scope determinants to answer the query; and send, to one or more scope determinants, a request to answer the query; and create, by means of the control language model, a response to the query based on one or more responses to the query received from one or more scope determinants. Side 12. The method according to Side 11, whereby each range-bound language model from the set of range-bound language models includes at least one neural network that has been trained with information about a different petrophysical domain and configured to answer queries related to the different petrophysical domain. Side 13. The method according to either Side 11 or 12, where each range-specific language model from the range-specific language model set includes at least one neural network that has been trained with different context-specific information and configured to answer queries associated with the different context-specific information, where the different context-specific information includes at least one tool-specific information and one well-hole-specific information. Aspect 14. The method according to any of the aspects from 11 to 13, where the computational language model is trained using petrophysical information, and each range-specific language model from among the set of range-specific language models can be trained using more specific petrophysical information compared to the computational language model. Aspect 15. The method according to any of the aspects from 11 to 14, where the query response is created based on the information received from one or more responses and the petrophysical information that the computational language model learned to create the response. Aspect 16. The method according to any of the aspects from 11 to 15, where one or more of the scope-bound language models include a subset of scope-bound language models from among the scope-bound language models set, and one or more of the responses include multiple responses from different scope-bound language models from the subset of scope-bound language models, and where the creation of the query response involves combining information from the multiple responses into a combined response. Aspect 17. The method according to any of the aspects from 11 to 16, where the control language model additionally includes a model configured to convert image data into textual data describing the image data, where the query includes visual information collected from the location of a well hole, and where at least one of the selection of one or more scoped language models and the creation of the response additionally depends on the visual information. Side 18. The method is in accordance with Side 17, where the visual information includes at least one petrophysical record, petrophysical map, petrophysical diagram, and petrophysical photograph. Aspect 19. The method according to any of the aspects from 11 to 18, where at least one of the control language model and one or more of the range-specific language model sets include a network of converters, and where the range-specific language model set includes at least one resistivity model set up to answer resistivity-related questions, a nuclear magnetic resonance (NMR) model set up to answer NMR-related questions, a porosity model set up to answer porosity-related questions, a fluid contact surface model set up to answer fluid contact surface-related questions, a permeability model set up to answer permeability-related questions, a configuration test model set up to answer configuration test-related questions, and a continuity model set up to answer continuity-related questions. Aspect 20. The method according to any of the aspects from 11 to 19, where one or more processors are further configured to do the following: based on the query, perform a relevant search for each of the one or more scalable language models, through a knowledge base containing petrophysical information; and generate, by the one or more scalable language models, one or more responses, based on each relevant search for each model in the knowledge base containing petrophysical information. Aspect 21. A non-temporary, computer-readable medium containing instructions which, when executed by one or more processors, cause one or more processors to execute a method according to any of Aspects 1 to 10. Aspect 22. A system that includes a means of implementing a method according to any of the aspects from 1 to 10. Aspect 23. A computer program product that includes instructions which, when executed by one or more computing devices, cause one or more computing devices to execute a method in accordance with any of Aspects 1 to 10.
Claims
1. A method that includes: receiving, by a control language model configured to perform natural language processing, a query relating to one or more subject areas; based on one or more subject areas related to the query and the respective domain-specific knowledge of each domain-specific language model from a set of domain-specific language models, selecting one or more domain-specific language models from the set of domain-specific language models to answer the query; sending, to one or more domain-specific language models, a request to answer the query; and generating, by the control language model, a response to the query based on one or more responses to the query received from one or more domain-specific language models.
2. The method according to safeguard element 1, where each domain-specific language model from the set of domain-specific language models includes at least one neural network trained with information about a different petrophysics domain and configured to answer queries related to that different petrophysics domain.
3. The method according to safeguard element 1, where each domain-specific language model from the set of domain-specific language models includes at least one neural network trained with different context-specific information and configured to answer queries related to that different context-specific information, where the different context-specific information includes at least one tool-specific piece of information and one well-hole piece of information.
4. The method according to safeguard element 1, where the control language model is trained using petrophysics information, and each domain-specific language model within the domain-specific language models is trained using petrophysics information that is more specific than that of the control language model.
5. The method according to safeguard element 1, where the query response can be generated based on the information from one or more responses and the petrophysics information learned by the control language model to generate the response.
6. The method according to safeguard element 1, where one or more domain-specific language models include a subset of domain-specific language models within the domain-specific language models.One or more responses include multiple responses from different domain-specific language models within the domain-specific language models subset, and the response generated for the query involves combining information from the multiple responses into a combined response.
7. The method according to safeguard element 1, where the control language model additionally includes a model configured to convert image data into text data describing the image data, where the query includes visual information collected from the wellbore site, and where at least one of the selections of one or more domain-specific language models and the response generated additionally relies on visual information.
8. The method according to safeguard element 7, where the visual information includes at least one petrophysical log, mapPetrophysical map, petrophysical chart, and petrophysical image.
9. The method according to safeguard element 1, where at least one control language model and one or more domain-specific language models are included on a transformer network, and where the domain-specific language models include at least one resistivity model configured to answer resistivity questions, a nuclear magnetic resonance (NMR) model configured to answer NMR questions, a porosity model configured to answer porosity questions, a fluid contact model configured to answer fluid contact surface questions, a permeability model configured to answer permeability questions, a formation testing model configured to answer formation testing questions, and a continuity model configured to answerQuestions related to continuity.
10. The method, according to safeguard element 1, also includes: performing a respective lookup, based on the query, by one or more domain-specific language models, in a knowledge base containing petrophysics information, and generating, by one or more domain-specific language models, one or more responses based on each respective lookup in the knowledge base containing petrophysics information.
11. The system includes: memory; and one or more processors associated with that memory, where one or more processors are configured to: receive, by a control language model configured to perform natural language processing, a query relating to one or more subject areas; based on one or more subject areasThe areas associated with the query and the respective domain-specific knowledge for each domain-specific language model from a set of domain-specific language models; selecting one or more domain-specific language models from the set of domain-specific language models to answer the query; sending a request to one or more domain-specific language models to answer the query; and generating, by the control language model, a response to the query based on one or more responses to the query received from one or more domain-specific language models.
12. The system, according to protection element 11, where each domain-specific language model from the set of domain-specific language models includes at least one neural network trained with information about a petrophysical domain.
13. The system, according to safeguard element 11, includes at least one neural network trained with different context-specific information and configured to answer queries related to that different context-specific information. This context-specific information includes at least one tool-specific piece of information and one wellbore-specific piece of information.
14. The system, according to safeguard element 11, includes a control language model trained using petrophysics information, and each domain-specific language model is trained fromAmong the domain-specific language models, using more specific petrophysics information compared to the control language model.
15. The system, according to safeguard element 11, where the query response is generated based on information from one or more responses and the petrophysics information learned by the control language model to generate the response.
16. The system, according to safeguard element 11, where one or more domain-specific language models include a subset of domain-specific language models from among the domain-specific language models, and one or more responses include multiple responses from different domain-specific language models within the subset of domain-specific language models, and where the response generation involvesThe query response combines information from multiple responses into a single, combined response.
17. The system, according to safeguard element 11, where the control language model includes, in addition, a model configured to convert image data into text data describing the image data, where the query includes visual information collected from a wellbore site, and where at least one of the selections of one or more domain-specific language models and the creation of the response are based on visual information, where the visual information includes at least one petrophysical log, petrophysical map, petrophysical chart, and petrophysical image.
18. The system, according to safeguard element 11, where at least one control language model and one or more domain-specific language models are included.Domain-specific language models on a transformer network, where the domain-specific language model set includes at least one resistivity model configured to answer resistivity questions, a nuclear magnetic resonance (NMR) model configured to answer NMR questions, a porosity model configured to answer porosity questions, a fluid contact model configured to answer fluid contact questions, a permeability model configured to answer permeability questions, a formation testing model configured to answer formation testing questions, and a continuity model configured to answer continuity questions.
19. The system, according to protection element 11, where one or more processors are also configured to perform a respective lookup, based on the query, by each domain-specific language model.One or more language models, in a knowledge base containing petrophysics information, and the generation, by one or more domain-specific language models, of one or more responses based on each respective lookup in the knowledge base containing petrophysics information.
20. A non-transitory computer-readable medium containing instructions which, when executed by one or more processors, cause one or more processors to: receive, by a control language model configured to perform natural language processing, a query related to one or more subject areas; based on the one or more subject areas associated with the query and the respective domain-specific knowledge for each domain-specific language modelFrom among a set of domain-specific language models, one or more domain-specific language models are selected from among a set of domain-specific language models to answer the query; a request is sent to one or more domain-specific language models to answer the query; and a response to the query is created, by a control language model, based on one or more responses to the query received from one or more domain-specific language models.
