Methods and systems for document validation and production

Abstract

A computerised method of analysing a document, e.g. for waybill validation. The method comprises receiving an image representation of a document, processing the image representation of the document us

Description

FIELD OF THE INVENTION This invention relates improvements in relation to cargo shipping processes and the validation of documents across a variety of transportation modes (including air, sea, rail, road, and other means). BACKGROUND OF THE INVENTION Conventionally, cargo shipment management relies heavily on manual preparation and validation of documentation such as air waybills, bills of lading, customs forms, and other regulatory paperwork. This reliance on paper-based or legacy systems frequently results in process inefficiencies, high labor costs, and an increased likelihood of human error. Moreover, regulations vary among different jurisdictions, carriers, and transportation modes, and often lack standardization. Shipments are therefore subject to delays, rejections, or even legal and financial penalties when documentation is incomplete or incorrect. Small and midsized businesses, or those shipping goods infrequently, find these requirements especially burdensome, as preparing documentation manually can be overwhelming and prone to missteps. The same challenges affect additional segments of the logistics sector, including warehousing operations and last-mile delivery. Recent advancements in Al, computer vision, and natural language processing (“NLP”)— particularly in the form of large language models—offer valuable opportunities to automate and enhance shipping workflows. Deep learning-based object detection can recognize cargo items and detect restricted items such as firearms or narcotics, while large language models can interpret text, images, or structured data to validate documents. Hybrid Al approaches, combining rule-based logic with neural network classifiers, can further strengthen compliance checks. However, to date, existing solutions do not provide holistic, end-to-end coverage that fully integrates multi-camera re-identification, RFID / Bluetooth-based tracking, and large language model-driven document validation across all transportation modes. SUMMARY OF THE INVENTION We have appreciated that there is an unmet need for an automated, unified system capable of creating and validating shipping documents, ensuring compliance with dynamic regulatory requirements, and enabling efficient cargo handling through Al and robotics. We have further appreciated that such a system would ideally incorporate automated detection of prohibited or suspicious items, generate required documentation (e.g., airwaybills, sea bills of lading, multimodal documents), perform comprehensive compliance checks, and interface with government or customs authorities to streamline cargo clearance. In broad terms, the invention provides a hybrid approach in which a document relating to shipping of goods may be validated by a combination of Al analysis and rules and I or in conjunction with automated document production. In this way, the invention leverages existing paper-based processes whilst improving accuracy and resilience. An embodiment of the invention provides improved methods and systems to handle authority transfer in cargo shipping, where multiple parties—including shippers, airlines, customs authorities, government agencies, sea carriers, road freight operators, and rail operators— must validate information, collaborate, and transfer knowledge or authority regarding goods. This invention relates an end-to-end automated cargo shipping processes and the validation of documents across a variety of transportation modes (including air, sea, rail, road, and other means). An embodiment of the invention leverages artificial intelligence (Al), computer vision, and robotics to streamline document validation and creation, ensure regulatory compliance, and facilitate automated cargo handling. By integrating Al-based validation, realtime data analysis, and robotic automation, the invention effectively addresses inefficiencies present in current manual processes and improves the accuracy and speed of shipping operations. A core feature of a embodiment of the invention is the use of Al models trained by collected item images and documents as well as large language models (LLMs), including multimodal variants, for advanced document reading in place of or alongside optical character recognition (“OCR”). This capability allows the system to detect incomplete forms, classify unknown items, and flag suspicious cargo (e.g., firearms, narcotics). By automating these processes, the system provides a robust approach to compliance that significantly reduces manual input errors and expedites shipping operations. In an embodiment, the invention contemplates utilizing data from existing paper-based processes while automating validation through Al and real-time data streams. Through advanced object detection, unknown items can be flagged for further inspection or quarantined. Robotic systems, guided by validated shipping information, can then pick, place, and track cargo autonomously. Governmental and customs authorities may verify shipments via a dedicated portal that leverages Al data and computer vision, thus enabling near-instant approvals and reducing the likelihood of fines, shipment delays, and other complications. Preferred features, as described in dependent claims, include integration of multiple cameras and CCTVs fortracking, RFID / Bluetooth tags, and re-identification algorithms for continuous, real-time cargo monitoring. The invention is defined by the independent claims, to which reference should now be made. Preferred features are laid out in the dependent claims. BRIEF DESCRIPTION OF THE DRAWINGS Embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which: Figure 1 shows the main components of a system and process of an embodiment of the invention relating to document validation; and Figure 2 shows the main components of a system and process of an embodiment of the invention relating to document production; Figure 3 shows an example user interaction for document relating to document validation; and Figure 4 shows an example system comprising components in accordance with an embodiment of the invention. DETAILED DESCRIPTION The following exemplary description is based on a system, apparatus, and method for use in the aviation industry. However, it will be appreciated that the invention may find application outside the aviation industry, particularly in other transportation industries such as sea transport, or delivery industries where cargo is handled and / or transported. The following embodiments described may be implemented using a Python programming language using for example an OpenCV library. However, this is exemplary and other programming languages known to the skilled person may be used such as Python and JAVA. An embodiment of the invention will be described in three aspects. In a first aspect, the embodiment involves validation of documents by a combination of Al and rules-based approaches. In a second aspect, automated document production is described. In the third, and main, aspect images of cargo items are combined with techniques of the first and second aspects to improve document production, validation and verification of cargo at points during transfer or at a destination. The first, second and third aspects may of course be used in combination with one another. Embodiments of the invention solve the problems discussed above by providing additional evaluation and validation of documentation that provides a transformative advantage by ensuring compliance ahead of time. Airlines benefit from streamlined operations, reduced liabilities, and faster acceptance of shipments, while shippers gain confidence in the accuracy of their submissions, reducing the risk of delays or penalties. Customs authorities also see enhanced efficiency through pre-validated shipments, expediting the clearance process and reducing workloads. Overall, the ecosystem benefits from improved reliability, reduced risks of asset seizure, and strengthened collaboration among shippers, airlines, and regulatory bodies. The system allows for a document to be validated to ensure compliance with country, International Air Transport Association (IATA), and / or airline-specific requirements. The system integrates a comprehensive knowledge base of country regulations, airline requirements, and embargo policies, dynamically updated via APIs to ensure that it can accurately validate information within a document according to current requirements. By leveraging optical character recognition (OCR) techniques and multimodal analysis, the system can extract shipment details and utilise artificial intelligence techniques to assess and fill out entries in a document (such as a shipping form). A validation module may also be implemented in order to check compliance of the document with cargo documentation requirement standards. A user interface may be provided that allows shippers and airline staff to interact with each other and provide step-by-step guidance for preparing and validating documents. A key feature of this system is flexibility to enable integration across platforms. For example, airlines may incorporate it into an airline-specific application or operational systems via APIs, while shippers may access it through a dedicated application or messaging platform. Such integration allows shipment details to be easily uploaded, validation feedback to be issued, and submitting a finalised document. Once a document is validated, the system may support electronic submission of the documents to airlines or regulatory authorities, thereby reducing processing times and ensuring smooth cargo handling. Optionally, a performance report may be generated to monitor validation accuracy and improve the system over time. By simplifying cargo documentation validation and submission, the system enhances efficiency, reduces risks, and strengthens the global supply chain. The system may also include features for scanning and validating Air WayBills (AWB) to ensure that all information is complete and compliant with Airway Regulations (RAG). This may be achieved by checking the for the presence of required documents while identifying and flagging irrelevant or unnecessary attachments, streamlining the submission process and reducing computational overhead for processing Airway requests. A compliance labeling system may provide immediate feedback. For example, shipments marked with a green label are compliant and ready to proceed, while a red label highlights missing documents, violations, or errors, such as TSA clearance omissions for the US, size restrictions for specific aircraft, or embargoes on restricted items. Figure 1 shows an example system 100 and associated functionalities configured to realise the technical advantages described herein. In the first aspect, in order to assess what information is necessary or to validate information about cargo, the system may comprise a database in the form of knowledge base 101 that can be interrogated and cross-referenced in response to a request. Such a request may be triggered by a user inputting a document to be validated. The knowledge base 101 may be a database and may be locally integrated into the system 100 or located remotely and be accessed by wired / wireless means. Stored within knowledge base 101 may be airline requirements 102, Embargo and Airway regulations 103, country regulations, and / or IATA documents 105. It will be appreciated that other information may be stored within knowledge base 101 to be referenced in response to a request. Knowledge base 101 may receive updates 106 to ensure that the regulations and IATA documents correspond to current standards. The updates 106 may be manually uploaded / downloaded or knowledge base 101 may be configured to check at regular intervals whether any updates are available. A user interface shown, here shown as user prompts 110, allows a user to provide input, referred to as a prompt, for example using a dedicated application or messaging platform. The user prompt contains information corresponding to one or more items of cargo to be transported. For example, said information may comprise a destination country 121 of the cargo, an airline 122 that the cargo will be transported on, a categorisation of the cargo (e.g., according to IATA cargo standards), and / or other relevant information. This information may be entered into the dedicated application or messaging platform by a user through a user interface. The system may also implement a chatbot 120 that prompts and guides a user to enter the requisite cargo information. As an example, the chatbot may issue a welcome message and invite the user to specify a shipping destination. The user may then respond to the message with the shipping destination (e.g., Munich, Germany) which can be used as the user prompt 110. The system can then request knowledge base 101 for Germany’s country regulations for cargo. After (or prior to) receiving information corresponding to the one or more items of cargo, the chatbot 120 may prompt the user to upload an image of one or more documents to be validated. It will be appreciated that an “image” of a document may also include a scan or otherwise captured photo of the document. The image may be received by a Large Language Model (LLM) Hub 130 that may comprise an optical character recognition (OCR) module 131 configured to performed OCR techniques to analyse the uploaded image to extract text (printed and / or handwritten). By way of example, the document comprised in the uploaded image may be an air waybill with one or more information fields to be completed based on information corresponding to the cargo. The air waybill may be incomplete, or all fields may be filled out by a user looking to confirm whether the form entries are correct. The OCR module 131 can perform OCR analysis to extract text from the airwaybill and input it into a large language model comprised in the LLM Hub 130 as selected by LLM selector 123. The LLM Hub 130 may then perform checking 133 to validate the text extracted from the air waybill, such as unseen data 140, as in this first aspect, and / or may fill up 132 a document to automatically populate fields in a document as will be discussed in relation to the second aspect below. The large language model contained in LLM Hub 130 can assess whether the extracted text corresponds correctly to requirements 161 of a given document in step 160. The model may be trained on a dataset that includes example text extracted from fields in correctly filled-out documents adhering to various regulatory requirements, such as those included in knowledge base 101. The LLM Hub 130 may interface with a RAG application programming interface (API) 142 to access and receive new regulations 141 as they are updated, ensuring that the most up-to-date requirements for documents are adhered to. The LLM output may be quality monitored 150 such as by using evaluation metrics (e.g., perplexity, accuracy and F1 score, human preference score), behavioural testing, A / B testing, and / or error analysis. The result from the quality monitoring may be returned to the LLM in a feedback loop to improve the model’s performance over time. Based on the output of the LLM, fields in the document may be completed at step 162 by a user performing user review 170 as in this first aspect, or text may be entered into the document fields automatically as in the second aspect below. This provides a validated document 163 that contains completed fields that meet regulatory requirements. A performance report 171 may be generated in the form of a confusion matrix with True positive (TP), False positive (FP), False negative (FN), and True negative (TN) values. The user may then upload the validated document 163 and submit the document in step 180 to a database of validated documents which can be interrogated by a cargo operator to review documentation. The database may also be accessible to regulatory authorities to review compliance with regulations. Figure 2 shows a system and process 200 of air waybill generation in accordance with a second aspect. In the second aspect, documents such as a airway bills may be produced automatically from data provided and guided by LLM models. The system 200 comprises an LLM Hub 230 which may correspond to LLM Hub 130 of figure 1 and its associated capabilities. A user may provide cargo information 220 to the system, for example through chatbot 120, that is relevant to completion of a document. For example, a user may input a cargo departure and destination location, weight, size, contents and other relevant cargo characteristics that may be subject to regulation. The LLM Hub 230 receives the user-provided information and can generate using a large language model, and based on the information, contextually appropriate text to be entered into a document such as an air waybill. The LLM Hub can reference knowledge base 201, which may correspond to knowledge base 101 of figure 1, and also interrogate RAG API 210, which may correspond to RAG API 142, to ensure the latest regulations are adhered to. The LLM may achieve this by leveraging language patterns, style, and contextual awareness derived from training on correctly populated document fields with labelled inputs. Hub 230 can perform steps 240 to generate a populated document. It will be appreciated that a populated document is a document in at least some of the fields contain entered information. The LLM Hub receives information provided by a user relating to the cargo and, based on this information, the LLM Hub can assess the requirements 241 that are imposed on the document to adhere to regulations. For instance, a user may input a cargo destination of Japan, and therefore Japanese importation regulations will be enforced. Based on the requirements determined in step 241, and the information provided by the user, the LLM can generate information and populate 242 one or more fields of a document using the generated information. LLM Hub 230 may store a database of documents with modifiable fields locally, or may interrogate a remote database of documents in which the generated information can be entered to retrieve a relevant document to be completed. Once the relevant field(s) in the relevant document have been populated, the document can be output 243 to a user for review and / or sent without further user input to a database of completed documents for regulatory review. We will now describe the third, and main, aspect of the embodiment of the invention. From the outset of the shipping process, each cargo item is photographed or scanned to create a detailed visual record. Such images may be input into the system as described in relation to figure 1. This initial imaging stage leverages Al-based object detection, which classifies items according to size, shape, and any visible labeling. The Al-based object detection may be operated by the processor shown as LLM Hub 130. Simultaneously, the system either assists in writing shipping documents—such as airwaybills, sea bills of lading, or multimodal transport forms—or automatically generates them by populating known fields (for example, cargo dimensions, declared contents, and destination). The large language model (LLM) 130 and knowledge base 101 then validate these newly produced documents to confirm that all mandatory information (including regulatory codes, handling instructions, and destinationspecific requirements) is present. LLMs help both verify shipping documents and generate new, regulation-compliant forms. For example, when a user uploads a partially completed air waybill or customs form, the LLM analyzes the text and images while also referencing a knowledge base of current regulations and embargo lists. This lets it spot missing fields or flag items subject to sanctions. Beyond validation, LLMs can fill in new forms automatically. If a user provides only basic information (e.g., item type, weight, destination), the LLM checks relevant rules such as IATA guidelines or embargo requirements and completes all necessary fields. It then adjusts the format for air, sea, or rail shipping and includes any special codes or permits for restricted materials. If an item is on a prohibited list, the LLM alerts users to remove it or provide the required documentation. Because LLMs also parse semi-structured data (like scanned certificates or cargo manifests), they can further confirm that listed goods match the shipping form. Tightly integrated with a live knowledge base, the model instantly adapts to new regulations or embargoes, ensuring that all forms and shipments stay up-to-date with changing rules. Once the items are officially documented, the same images and associated shipping records move through the subsequent transport stages. Whether the cargo is traveling by air, sea, rail, or road, multimodal LLMs and real-time data streams keep the shipment synchronized with its respective documents. At warehouses and checkpoints, multi-camera systems or handheld devices cross-reference the physical items—using computer vision and reidentification algorithms—with the previously captured images and verified paperwork. Again, such images may be input into the system shown in figure 1 and validated by the LLM Hub 130. If the system detects discrepancies, such as a mismatched item or a newly discovered prohibited good, it alerts the relevant parties (e.g., customs officials or warehouse operators) to take corrective action immediately. As the cargo reaches its final destination, customs authorities have the option to review the validated documents and compare them against the original item images through a secure government portal. Al-driven checks ensure that every piece of cargo matches the information declared in the shipping forms and that no required labels or special clearances have been overlooked. Any irregularities—such as missing permits, unrecognized items, or hazards—are flagged automatically for further scrutiny. Conversely, if all fields and images are in order, the customs clearance process becomes straightforward and efficient. By using a unified set of item images, updated regulations, and cross-verified documents, the system minimizes manual interventions while maintaining a high level of accuracy and compliance throughout the entire shipping journey. A curated dataset of annotated air waybills, customs forms, embargo lists, and cargo item images helps train the model on real-world logistics data. Exposing the LLM to domainspecific terms (e.g., product codes, restricted-goods identifiers) and visual cues (e.g., barcodes, container types) enables it to detect missing fields, parse specialized regulations, and accurately recognize different items. This targeted fine-tuning also addresses regionspecific documentation rules and niche cargo categories, boosting its ability to auto-fill and validate shipping forms. As a result, the refined model more consistently flags prohibited materials, aligns documents with actual cargo, and keeps shipments compliant across diverse global routes. Based on the operations of the LLM Hub 130, the chatbot 120 may output an analysis of the document in the uploaded image. Said analysis may comprise one or more requirements 160 that have been identified by the LLM Hub. A requirement may correspond to an identified information field in the document that is either incomplete or incorrect and therefore requires completion or correction. The term “information field” may correspond to an input area in the document where a user can enter information corresponding to the cargo. For example, the form may include a “Weight” field which in which the weight of the cargo can be entered. Figure 3 shows an example chatbot interaction 300. A user can input information 310, for example about the cargo to be transported, which can act as user prompt 110 of figure 1. In this example, the user has established that the document is an air waybill and the relating cargo is to be shipped to Japan. As discussed above, this information can be used by an LLM Hub to ensure that the relevant regulations are met for cargo documentation. In this example, a user uploads an image representation of document 320 (e.g, a scan or photograph) that has been populated for validation by the system in accordance with the first aspect discussed herein. Using the techniques described herein, it can be established whether the fields in document 320 have been populated correctly, and a validation result 330 can be output either approving the document or highlighting deficiencies to be corrected. In this example, it has been determined that the “phytosanitary form” is missing and therefore validation has failed. A user can then upload a second image representation of the document 320 that has been adjusted to correct the deficiencies (if any are identified). The second image representation may also be checked to ensure that the document fields are correctly populated using. After uploading the second image representation, the chatbot can output a further validation result. If this is a positive result, the user can then submit 340 the second image representation to a database of document images for access by cargo handlers and / or regulatory bodies. In this instance, the submission can be triggered by typing “Submit”, or by clicking a submit option in the chatbot interface. In the event that the second validation result is negative, the process above may be repeated until a positive validation result is obtained. It will be appreciated that figure 3 shows an example chatbot interaction in accordance with the first aspect discussed herein. However, the chatbot may additionally or alternatively be configured to perform in accordance with the second aspect discussed herein. In this case, no image representation 320 is supplied by the user, and instead the chatbot outputs a populated document after input information 310 is supplied. The user can then submit 340 the populated document as described above. Figure 4 shows a distributed system 400 in accordance with an embodiment of the invention. Each node in this system may interact via wireless or wired transmission and / or may be comprised in the same node. The system may comprise user device 410 which can provide user prompt 110, cargo information 220, information 310, image representation 320, and / or document submission 340. Device 410 may be a mobile device such as a smartphone or tablet configured to generate an image representation of a document, for example using a camera. The device may also store and run software configured to enable chatbot interaction such as that of figure 3. Co-ordinating node 420 may comprise an artificial intelligence model to provide chatbot functionality on device 410. This model may leverage natural language processing (NLP) techniques to generate instructions to a user of device 410 and to also interact with LLM 430 which is configured to perform the functions of LLM Hub 130. In this way, node 420 can operate as a front-end interface that communicates with the LLM 430. Node 420 may also transmit and receive information from Application Programming Interface (API) 440 including validated documents which can be stored in a database 441. For example, the submission 180 of figure 1 may be received by API 440 and stored in database 441. API 440 can then act as a communication layer between the database 441 and dashboard 450 and / or airline interface 460 to send, retrieve, update, or delete information in database 441. Dashboard 450 may be software used, for example, by a regulatory body to access and review documentation stored in the database to check whether it complies with regulatory requirements. Similarly, an airline may use airline interface 460 to access and review documentation to check whether it meets a set of requirements. Exemplary embodiments of the invention may be implemented as a circuit board which may include a CPU, a bus, RAM, flash memory, one or more ports for operation of connected I / O apparatus such as printers, display, keypads, sensors and cameras, ROM, and the like. The wired or wireless communication networks described above may be public, private, wired or wireless network. The communications network may include one or more of a local area network (LAN), a wide area network (WAN), the Internet, a mobile telephony communication system, or a satellite communication system. The communications network may comprise any suitable infrastructure, including copper cables, optical cables or fibres, routers, firewalls, switches, gateway computers and edge servers. The system described above may comprise a Graphical User Interface. Embodiments of the invention may include an on-screen graphical user interface. The user interface may be provided, for example, in the form of a widget embedded in a web site, as an application for a device, or on a dedicated landing web page. Computer readable program instructions for implementing the graphical user interface may be downloaded to the client device from a computer readable storage medium via a network, for example, the Internet, a local area network (LAN), a wide area network (WAN) and / or a wireless network. The instructions may be stored in a computer readable storage medium within the client device. As will be appreciated by one of skill in the art, the invention described herein may be embodied in whole or in part as a method, a data processing system, or a computer program product including computer readable instructions. Accordingly, the invention may take the form of an entirely hardware embodiment or an embodiment combining software, hardware and any other suitable approach or apparatus. The computer readable program instructions may be stored on a non-transitory, tangible computer readable medium. The computer readable storage medium may include one or more of an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk. The above detailed description of embodiments of the invention is not intended to be exhaustive or to limit the invention to the precise form disclosed. For example, while processes or blocks are presented in a given order, alternative embodiments may perform routines having steps, or employ systems having blocks, in a different order, and some processes or blocks may be deleted, moved, added, subdivided, combined, and / or modified. Each of these processes or blocks may be implemented in a variety of different ways. Also, while processes or blocks are at times shown as being performed in series, these processes or blocks may instead be performed in parallel or may be performed at different times. The teachings of the invention provided herein can be applied to other systems, not necessarily the system described above. The elements and acts of the various embodiments described above can be combined to provide further embodiments. While some embodiments of the inventions have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the disclosure. Indeed, the novel methods and systems described herein may be embodied in 5 a variety of other forms; furthermore, various omissions, substitutions and changes in the form of the methods and systems described herein may be made without departing from the spirit of the disclosure.

Claims

1. A computerised method of analysing a document to provide validation of document contents, comprising:- receiving an image representation of a document;- processing the image representation of the document using a large language model process to provide an output indicating one or more fields within the document;- processing the output using a set of rules to determine whether the one or more fields satisfy the set of rules; and- generating a message according to the determination of whether the one or more fields satisfy the set of rules.

2. The method of claim 1, wherein the large language model is trained using example documents having separate areas for each field.

3. The method of claim 1 or 2, wherein the document is a waybill specifying goods for transport.

4. A computerised method of producing a document, comprising:- receiving an image of goods for shipping;- processing the image of the goods using a large language model process to provide an output indicating one or more categories;- processing the output using a set of rules to determine whether the one or more categories satisfy the set of rules; and- generating a document according to the determination of whether the one or more categories satisfy the set of rules.

5. The method of claim 4, wherein the large language model is trained using example documents having separate areas for each field.

6. The method of claim 4 or 5, wherein the document is a waybill specifying goods for transport.

7. A computerised method of analysing images of transported goods in relation to an accompanying a document to provide validation of the goods, comprising:- receiving an image of the goods;- processing the image of the goods using a large language model process to provide an output indicating whether the goods match the accompanying document;- processing the output using a set of rules to determine whether the goods and accompanying document satisfy the set of rules; and5 - generating a message according to the determination of whether the goods andaccompanying document satisfy the set of rules.A