AI-based inventory optimization system for pharmaceutical supply chains

Abstract

An AI-based inventory optimization system implemented as a device for pharmaceutical supply chains. The system includes: a structural frame defining a plurality of modular storage compartments, each compartment equipped with robotic racks mounted on motorized drive assemblies, the drive assemblies configured to translate drug containers along three-dimensionally guided rails; an array of environmental sensors arranged in the compartments, the sensors comprising temperature sensors, humidity sensors, and vibration sensors, each sensor electrically connected to an edge processing module housed in the frame, the edge processing module configured to pre-process sensor data using convolutional neural network models to detect deviations in cold chain compliance; a central processing unit comprising at least one GPU or neural processing accelerator, the central processing unit executing a recurrent neural network-based demand forecasting engine and a reinforcement learning agent-based optimization engine, the processing unit generating control signals for redistribution policies that minimize predicted shortages, waste, and regulatory violations; a conveyor-driven dispensing unit operatively connected to the actuator units, the dispensing unit comprising belt actuators, RFID tag readers, and load sensors, and configured to redistribute goods between compartments or to external dispensing points; a communications gateway that supports both 5G and LPWAN connectivity protocols for real-time synchronization with distributed supply chain nodes and cloud servers; and A blockchain-integrated cryptographic module embedded in the central processing unit, configured to digitally sign and record every inventory movement, environmental condition, and redistribution command in a distributed ledger, thus ensuring immutable traceability of the drug inventory.

Description

Field of the invention

[0001] The present invention relates to pharmaceutical logistics and supply chain management, and more particularly, to an AI-based inventory optimization system that integrates real-time demand forecasting, machine learning-based inventory allocation, and IoT-enabled monitoring to minimize shortages, reduce waste of temperature-sensitive medications, and ensure regulatory compliance. The invention also relates to a device in the form of a machine structure with computing modules, sensor integration hardware, and actuator-controlled dispensing and storage control mechanisms for automated pharmaceutical inventory optimization across distributed supply chain nodes. Background of the invention

[0002] Pharmaceutical supply chains face the significant challenge of meeting demand while avoiding overstocking, especially for temperature-sensitive products such as vaccines, biologics, and narcotics. Existing inventory management systems are often rule-based and rely on traditional demand forecasting methods such as exponential smoothing or seasonal ARIMA models. These solutions cannot dynamically adapt to sudden fluctuations in demand due to epidemics, political changes, or production disruptions. Furthermore, they typically operate in silos, with minimal integration between distributors, wholesalers, and pharmacies. This can lead to critical medications being overstocked at one hub while simultaneously experiencing shortages at another.

[0003] Traditional warehouse management systems rely heavily on human input and static safety stock calculations, which are error-prone and inefficient. While enterprise resource planning (ERP) systems are available, they are typically unable to process heterogeneous data streams such as IoT sensor logs, environmental monitoring, shipment telemetry, and patient prescription data. Furthermore, pharmaceutical supply chains are subject to stringent regulations and require strict adherence to temperature, humidity, and traceability requirements. Existing solutions do not adequately integrate predictive analytics with real-time actions, such as automatic inventory reallocation or adaptive allocation based on demand fluctuations.

[0004] Therefore, there is a need for an intelligent system that can not only forecast drug demand but also dynamically optimize inventory distribution across multiple nodes while ensuring cold chain integrity. Such a system must also be integrated into a device or machine architecture that integrates AI computations, edge IoT data processing, cloud-based learning modules, and physical control mechanisms to perform real-time, self-optimizing inventory adjustments.

[0005] The pharmaceutical industry operates under highly complex supply chain conditions, ranging from raw material suppliers and manufacturers to wholesalers and distributors, hospitals and pharmacies. Ensuring the availability of vital medicines, vaccines, and other pharmaceutical products at the right time and in the right place is critical to protecting public health, but remains one of the most challenging logistical areas. The importance of balancing supply and demand is further emphasized by the fact that many medicines are temperature-sensitive, have a limited shelf life, and must comply with strict regulatory guidelines.Despite significant advances in enterprise resource planning and logistics management technologies, inventory management in the pharmaceutical industry continues to face persistent problems such as stockouts, overstocking, expiration-related waste, and cold chain violations. Existing solutions attempt to address these challenges through traditional forecasting models, manual interventions, and digitized enterprise platforms, but often fail to provide resilient and adaptive optimization in the face of volatile demand patterns and global disruptions.

[0006] Traditional pharmaceutical supply chains are primarily based on static safety stock methods, where each node in the chain maintains a predefined buffer stock based on historical sales data and lead times. While this approach provides a degree of protection against shortages, it is highly inefficient under conditions of fluctuating demand or uncertain supply. For example, during a seasonal flu outbreak or an unexpected epidemic, demand for certain antiviral drugs or vaccines may surge unpredictably, rendering safety stock calculations invalid. Conversely, if demand is overestimated, excess stocks may remain unused and expire, leading to financial losses and the waste of critical medical resources. Traditional models cannot dynamically update their parameters based on real-time conditions.This leads to inefficiencies that have a cascading impact on the entire supply chain.

[0007] Another widely used approach is the use of classical statistical forecasting techniques such as moving averages, exponential smoothing, and autoregressive integrated moving average (ARIMA) models. These methods can capture seasonal trends and fundamental demand cycles, but they do not perform well in the face of sudden structural changes such as politically motivated vaccination campaigns, recalls of contaminated batches, or supply disruptions due to geopolitical or production-related crises. Because they rely on steady-state assumptions, their adaptability to rapidly changing environments is limited. Furthermore, they cannot account for heterogeneous data sources such as patient-level prescription trends, IoT sensor readings from warehouses, or external signals such as disease surveillance data, which could otherwise improve forecast accuracy.

[0008] The enterprise resource planning (ERP) systems commonly used in pharmaceutical companies combine procurement, warehousing, and distribution processes on a unified platform. While these systems centralize data management and offer standardized processes, they are not designed for highly dynamic inventory optimization across distributed nodes. ERP solutions are often based on pre-programmed business rules and human decision-making rather than autonomous, learning-based mechanisms.

[0009] Decisions regarding replenishment, reallocation, and adjustment of safety stocks are therefore reactive rather than proactive, leading to delayed reactions to fluctuations in demand. Furthermore, ERP systems are often rigid and require extensive adaptation to pharmaceutical-specific compliance requirements such as temperature-controlled storage and batch-level traceability, making implementation more complex and expensive.

[0010] More recently, supply chain management systems have experimented with demand-driven replenishment (DDR) and vendor-managed inventory (VMI). In DDR systems, inventory is replenished based on actual consumption rather than forecasted demand, helping to reduce stockouts. In VMI systems, suppliers assume responsibility for maintaining optimal inventory levels at the customer's site. While both approaches shift some responsibility from distributors to suppliers and enable closer coordination, they suffer from critical limitations. Because DDR focuses on consumption, it may not anticipate sudden spikes in demand and often fails to respond adequately to spikes in demand.VMI, on the other hand, requires suppliers to have sufficient visibility across all demand nodes, which is rarely the case in fragmented pharmaceutical networks where wholesalers and distributors supply multiple independent pharmacies. Both methods still require sophisticated forecasting and coordination mechanisms to function effectively, which many current implementations lack.

[0011] Cold chain management represents another major bottleneck in pharmaceutical inventory optimization. Vaccines, biologics, and certain injectable formulations must be stored within strict temperature ranges. Existing cold chain monitoring systems are often isolated and use standalone IoT temperature loggers that alert in the event of deviations. However, these systems are typically not integrated with inventory management platforms. This means that while a breach may be detected, it does not automatically trigger intelligent redistribution or reallocation strategies to prevent at-risk inventory from entering circulation. This lack of integration between environmental monitoring and inventory optimization exposes supply chains to compliance violations, financial loss, and potential patient harm.

[0012] Another emerging solution is blockchain-based traceability platforms designed to combat drug counterfeiting and increase transparency in supply chains. While blockchain provides an immutable ledger for tracking pharmaceutical products, it often focuses more on traceability than optimization. While they ensure authenticity and provenance, they don't necessarily solve the problems of inventory imbalances, overstocking, or dynamic redistribution. Blockchain solutions are often layered on top of existing ERP systems, creating additional complexity without addressing fundamental inefficiencies in inventory allocation.

[0013] Artificial intelligence and machine learning are increasingly being incorporated into pharmaceutical supply chain research, but current implementations are often limited. Many pilot projects use machine learning models in isolation for demand forecasting, without integrating them into a closed-loop optimization system. For example, a predictive model may anticipate higher vaccine demand in a specific region, but lack the integration with distribution or inventory movement systems to translate this forecast into concrete supply adjustments. This disconnect between predictive analytics and actionable inventory optimization represents a major shortcoming of existing AI-powered initiatives.

[0014] Furthermore, current pharmaceutical inventory systems often lack real-time interoperability between stakeholders. Data is fragmented across hospitals, pharmacies, distributors, and manufacturers, making it difficult to obtain a holistic view of supply and demand. This fragmentation is further exacerbated by differences in electronic health records (EHRs), prescription databases, and regional regulatory requirements. Without harmonized data streams, even the most advanced forecasting techniques struggle to achieve high levels of accuracy. Latency in information exchange can also lead to delays in decision-making, resulting in shortages or overstocks.

[0015] Another significant disadvantage of existing solutions is the lack of autonomous physical execution mechanisms. Most inventory optimization platforms operate at the decision support level, providing reports or dashboards to human managers, who must then manually implement reallocation or replenishment strategies. This human dependence slows response times and introduces subjectivity, which is particularly problematic in pharmaceutical supply chains, where timing and accuracy are critical. During the COVID-19 pandemic, for example, vaccine distribution experienced significant delays, partly because manual allocation decisions could not keep pace with rapidly changing infection patterns and demand spikes.A system that merely provides insights without being directly integrated with mechanical actuators or robotic storage systems cannot fully solve the challenge of real-time inventory optimization.

[0016] Scalability is another unresolved limitation of current systems. While small-scale pilot projects can enable improved demand forecasting or inventory management, scaling these solutions across national or global pharmaceutical networks presents significant computational, infrastructure, and regulatory challenges. Legacy IT systems in many hospitals and pharmacies may not support integration with modern platforms, and compliance with diverse international regulations complicates global deployment. Existing solutions lack the adaptability to function seamlessly across diverse infrastructure environments while ensuring compliance and reliability.

[0017] Finally, most existing solutions narrowly focus on cost reduction or inventory efficiency without explicitly balancing the three objectives of the pharmaceutical supply chain—patient safety, regulatory compliance, and operational efficiency. An overemphasis on cost optimization can inadvertently increase the risk of shortages of life-saving medicines, while a sole focus on regulatory compliance can lead to overstocking and waste. The lack of systems that simultaneously optimize these competing objectives reflects a fundamental gap in current technological approaches.

[0018] These limitations highlight that existing inventory management systems in the pharmaceutical industry remain fragmented, reactive, and inefficient. They are unable to combine predictive intelligence, real-time adaptability, autonomous control, and end-to-end compliance into a unified framework. The challenges of heterogeneous data integration, demand volatility, cold chain monitoring, regulatory traceability, and operational scalability require a next-generation solution that is not only software-driven but embodied as an integrated AI-enabled device with closed-loop optimization. The inadequacies of existing methods form the technical basis and necessity for the proposed invention. Summary of the invention

[0019] The invention describes an AI-driven system and device for optimizing pharmaceutical inventory levels. The system comprises a multi-layered architecture with data acquisition modules, AI-based forecasting engines, optimization solvers, and control interfaces. Data from pharmacy sales records, hospital prescription systems, vendor logs, and IoT-enabled environmental monitors are compiled into a common format and transmitted to a central AI computation module. This module utilizes deep learning models, such as LSTM-based recurrent neural networks, for temporal demand forecasting and reinforcement learning agents for dynamic inventory reallocation policies.

[0020] The device design comprises a structural enclosure that integrates (a) an AI processor module with embedded GPUs or NPUs, (b) IoT sensor arrays for measuring temperature, humidity, vibration, and inventory levels, (c) robotic conveyor racks with motorized drives for retrieving and redistributing medication packages, and (d) a secure communications gateway connected to a blockchain ledger to ensure end-to-end traceability. The AI ​​computing module is directly connected to the drive controllers to trigger the physical redistribution of inventory within warehouses or between logistics vehicles. This combination of predictive intelligence and mechanical actuation represents a closed-loop inventory optimization device specifically tailored for pharmaceutical supply chains.

[0021] The primary objective of the present invention is to provide a highly intelligent and adaptive system for inventory optimization in pharmaceutical supply chains capable of continuously analyzing dynamic patterns of demand, supply, and environmental conditions and autonomously implementing optimal inventory management strategies. Unlike traditional rule-based or manual approaches, the invention aims to establish a self-learning mechanism that evolves over time using advanced artificial intelligence models. This ensures that medicines and healthcare products are always available in the right place and in the right quantity, avoiding excessive overstocking or shortages.The system is designed to directly address the challenges of unpredictable demand, the perishability of temperature-sensitive pharmaceuticals, and the rigid framework of existing enterprise solutions.

[0022] Another objective of the invention is to create a unified framework that bridges the gap between predictive analytics and actionable implementation in inventory management. While existing solutions provide decision-supporting results in the form of reports or dashboards, they still require human intervention to implement these insights. The present invention, however, integrates AI-based forecasting engines with actuator-controlled storage and issue mechanisms, thus closing the loop from data collection to physical action. This ensures that real-time forecasts of demand or inventory imbalances can be immediately translated into automated reallocation, replenishment, or re-allocation actions in warehouses, distribution centers, and pharmacies. By embedding intelligence into the computational and mechanical domains, the system achieves a level of autonomy not offered by current solutions.

[0023] Another goal of the invention is to comply with strict pharmaceutical regulatory standards while optimizing inventory management. The system enables real-time monitoring of storage conditions such as temperature, humidity, and vibration, ensuring that cold chain requirements are always met. In case of deviations, the AI ​​module can initiate corrective actions, such as redistributing inventory to compliant storage compartments, activating embedded cooling subsystems, or triggering an alert to higher-level regulatory systems. The invention thus ensures that inventory optimization does not come at the expense of patient safety or product quality, but rather strengthens the integrity and traceability of pharmaceutical products from the manufacturer to the end user.

[0024] Another goal of the invention is the interoperability and harmonization of the fragmented data systems currently existing in pharmaceutical supply networks. By integrating data streams from disparate sources such as hospital prescription systems, pharmacy sales records, distributor logs, IoT sensor data, and blockchain-based traceability ledgers, the system creates a unified, real-time view of the entire supply chain. This harmonized data pool forms the basis for more precise forecasting and optimization, overcoming the current challenges of siloed data management and inconsistent information sharing. The invention thus promotes a collaborative ecosystem in which manufacturers, distributors, hospitals, and pharmacies can operate on a common intelligent platform.

[0025] Another objective of the invention is to reduce the immense financial burden caused by the waste of expired or improperly stored pharmaceuticals. Current systems often fail to prevent overstocking or fail to respond appropriately when storage conditions exceed limits, resulting in significant product losses. The AI-based system proposed in this invention predicts expiration dates in advance, dynamically distributes near-expiration products to regions of higher consumption, and integrates mechanical dispensing systems to prioritize stock rotation based on shelf life. This predictive waste reduction mechanism not only increases economic efficiency but also contributes to more sustainable supply chain practices by reducing unnecessary pharmaceutical disposal.

[0026] Another goal of the invention is to improve the resilience of pharmaceutical supply chains in crises such as pandemics, natural disasters, or geopolitical upheavals. In such scenarios, conventional static models and manual coordination mechanisms cannot keep pace with rapidly evolving demand and distribution challenges. The new reinforcement learning framework continuously adapts to new conditions, simulates different distribution strategies, and autonomously selects the optimal course of action. In combination with actor-based redistribution systems, this capability enables the supply chain to respond flexibly and precisely even under extreme uncertainty, thus ensuring continuous access to life-saving medicines.

[0027] Another objective of the invention is to develop a device design that physically combines AI computing modules, IoT-based sensor arrays, robotic storage racks, drive-controlled conveyor belts, and blockchain-enabled communication gateways into a single integrated structure. By designing the system not just as software, but as a tangible machine architecture, the invention ensures that computational intelligence is seamlessly linked to the physical design. This hardware-software convergence enables the system to be deployed in real-world warehouse environments, distribution centers, and even pharmacies, where it can optimize storage, replenishment, and shipping processes autonomously and without constant human supervision.

[0028] Another goal of the invention is to improve patient safety and healthcare by ensuring uninterrupted access to critical medicines while minimizing the risk of counterfeit or compromised medications. The system achieves this through a blockchain-integrated traceability module that logs every inventory movement, environmental condition, and redistribution in an immutable ledger. This not only prevents counterfeiting but also provides regulators, manufacturers, and healthcare providers with complete transparency over the pharmaceutical product lifecycle. By combining optimization, compliance, and security, the invention meets both logistical and public health goals.

[0029] Another objective of the invention is to achieve scalability and adaptability across different geographic regions, regulatory frameworks, and infrastructure conditions. Many existing systems are limited by outdated IT architectures or region-specific compliance frameworks, which restrict their global scalability. However, the present invention leverages a hybrid edge-cloud architecture that enables real-time processing at local nodes while coordinating more comprehensive optimization strategies at the cloud level. This ensures that even regions with limited connectivity or infrastructure benefit from localized intelligence while contributing to and meeting global supply chain optimization goals. SHORT DESCRIPTION OF THE FIGURE

[0030] These and other features, aspects, and advantages of the present invention will become more readily understood when the following detailed description is read in conjunction with the accompanying drawings, in which like characters represent like parts throughout. Fig. Figure 1 shows a block diagram of an AI-based inventory optimization system for pharmaceutical supply chains.

[0031] Those skilled in the art will also appreciate that the elements in the drawings are shown for convenience and are not necessarily to scale. For example, the flowcharts illustrate the method by key steps to enhance understanding of aspects of the present disclosure. Moreover, with respect to device construction, one or more components of the device may be represented in the drawings by conventional symbols, and the drawing may show only the specific details relevant to understanding embodiments of the present disclosure in order not to clutter the drawings with details that would be readily apparent to those skilled in the art after reading the present description. Detailed description of the invention

[0032] For a better understanding of the principles of the invention, reference is made below to the embodiment illustrated in the drawings and described in specific language. However, the scope of the invention is not limited thereby. Changes and further modifications to the illustrated system, as well as further applications of the principles of the invention, are possible, as would normally occur to one skilled in the art to which the invention pertains.

[0033] It will be understood by those skilled in the art that the foregoing general description and the following detailed description are exemplary and explanatory of the invention and are not intended to be limiting thereof.

[0034] References in this specification to "one aspect," "another aspect," or similar expressions mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present disclosure. Therefore, the occurrences of the terms "in one embodiment," "in another embodiment," and similar expressions throughout this specification may or may not all refer to the same embodiment.

[0035] The terms "comprises," "having," or other variations thereof are intended to cover non-exclusive inclusion, such that a process or method comprising a list of steps not only includes those steps, but may also include other steps not expressly listed or inherent in such process or method. Likewise, the statement "comprises" with respect to one or more devices, subsystems, elements, structures, or components does not exclude, without further limitation, the existence of other devices, other subsystems, elements, structures, or components, or additional devices, additional subsystems, additional elements, additional structures, or additional components.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the invention pertains. The system, methods, and examples provided herein are for illustrative purposes only and should not be considered limiting.

[0037] Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0038] Figure 1 shows a block diagram of an AI-based inventory optimization system for pharmaceutical supply chains. The system 100 comprises: a structural frame (102) defining a plurality of modular storage compartments (102a), each compartment equipped with robotic racks (102b) mounted on motorized drive assemblies, the drive assemblies configured to move drug containers along three-dimensionally guided rails; an array of environmental sensors (104) arranged in the compartments, including temperature sensors, humidity sensors, and vibration sensors, each sensor electrically connected to an edge processing module housed in the frame, the edge processing module (104a) configured to pre-process sensor data using convolutional neural network models to detect deviations in cold chain compliance;a central processing unit (106) comprising at least one GPU or neural processing accelerator, the central processing unit executing a demand forecasting engine based on recurrent neural networks and an optimization engine based on reinforcement learning agents, the processing unit generating control signals for redistribution policies that minimize predicted shortages, waste, and regulatory violations; a conveyor-driven dispensing unit (108) operatively connected to the actuator units, the dispensing unit comprising belt actuators, RFID tag readers, and load sensors and configured to redistribute inventory between bins or to external dispensing points; a communications gateway (110) supporting both 5G and LPWAN connectivity protocols for real-time synchronization with distributed supply chain nodes and cloud servers;and a blockchain-integrated cryptographic module (112) embedded in the central processing unit, configured to digitally sign and record each inventory movement, environmental condition, and redistribution command in a distributed ledger, thus ensuring immutable traceability of the drug inventory;

[0039] In one embodiment, the robotic shelves (102b) are constructed as modular panels mounted along vertical and horizontal guide rails, with each shelf driven by servo motors controlled by pulse width modulation signals, the servo motors further equipped with position sensors to provide real-time feedback to the central processing unit, and the shelves configured to selectively retrieve, relocate, and redeploy drug containers under AI-generated control policies that adaptively reallocate inventory to high- and low-demand compartments in response to predicted consumption trends.

[0040] In one embodiment, the environmental sensors (104) are networked via a local bus architecture connecting each compartment to the edge processing module, wherein the edge processing module performs simple anomaly detection techniques that identify deviations in temperature gradients or sudden vibration events, and wherein the module is further configured to issue corrective actuation signals that either activate Peltier-based thermoelectric cooling subsystems, start compartmentalized circulation fans, or trigger the relocation of goods from a compromised compartment to an adjacent compatible compartment, thereby maintaining the integrity of the cold chain in real time.

[0041] In one embodiment, the conveyor-driven dispensing assembly (108) comprises a plurality of belt drives arranged in parallel channels, each drive mechanically connected to a motor with torque control feedback, the belts equipped with embedded RFID tag readers and load cell sensors configured to verify the authenticity and quantity of the transferred drug packages, and the assembly further comprising a mechanical diverter arm driven by a stepper motor, the diverter arm being controlled by the central processing unit to direct drug units either to shipping modules for external delivery or to redistribution channels for redistribution within the device.

[0042] In one embodiment, the communications gateway (110) comprises a dual-band antenna array embedded in the device housing. The gateway is configured to operate simultaneously on high-throughput 5G networks for cloud synchronization of AI model updates and on low-power LPWAN networks for continuous telemetry reporting to pharmacies in rural areas or with limited connectivity. The gateway also includes an adaptive routing module that dynamically selects the communications protocol based on real-time bandwidth availability and latency thresholds, ensuring uninterrupted connectivity across diverse supply chain environments.

[0043] In one embodiment, the blockchain-integrated cryptographic module (112) comprises a hardware-secured coprocessor implementing elliptic curve digital signature techniques. The coprocessor is configured to generate unique digital fingerprints for each drug movement transaction. These fingerprints are linked to environmental compliance metadata and transmitted to a distributed blockchain ledger via the communication gateway. The ledger ensures that regulators and supply chain stakeholders can audit an immutable record of each drug unit's storage history, compliance status, and distribution channel.

[0044] In one embodiment, the central processing unit (106) executes a reinforcement learning-based optimization engine configured to simulate multiple inventory reallocation scenarios in parallel. The engine is implemented on GPU hardware to accelerate policy evaluation across stochastic demand inputs. The optimization engine computes a multi-objective cost function that considers the probability of out-of-stocks, the risk of expiration losses, and the probability of compliance deviations, selects an optimal reallocation policy, and translates the policy into actuator-level control signals for robotic shelves, conveyors, and shelf actuators. This ensures that the optimization results are immediately incorporated into a physical reallocation within the device.

[0045] In one embodiment, the modular storage compartments (102a) are further divided into dynamically reconfigurable sections, each section being defined by actuator-controlled sliding barriers driven by linear actuators guided on precision rails, and wherein the central processing unit actuates the barriers to change the size of the compartment volumes in response to predicted fluctuations in demand, thereby enabling adaptive partitioning of storage capacity into high, medium, and low demand areas without requiring physical reassembly of the device structure.

[0046] In one embodiment, the edge processing module (104a) and the central processing unit (106) operate in a hybrid cloud-edge configuration, wherein the edge processing module is configured to perform near-real-time anomaly detection and remediation actions in conditions of limited connectivity, while the central processing unit offloads aggregated data to cloud-based servers to periodically retrain the recurrent neural network prediction models. The updated models are redistributed to the device via the communication gateway, ensuring continuous learning and adaptation without interrupting local operational autonomy.

[0047] In one embodiment, the structural frame (102) of the device comprises a thermally insulated composite enclosure, wherein Peltier-based thermoelectric cooling plates and distributed thermal sensors are embedded within each wall segment. The enclosure further incorporates a closed-loop air circulation system comprising brushless DC fans and ductwork. The circulation system is actuated by the central processing unit in response to sensor data to maintain a uniform temperature across all compartments, ensuring full compliance with pharmaceutical cold chain requirements while supporting mechanical inventory movement.

[0048] The present invention describes an AI-based inventory optimization system as a physical device for pharmaceutical supply chains. Both the mechanical structure and computational intelligence are integrated into a unified, closed-loop machine. The system is based on a modular structural frame that defines multiple storage compartments. Each compartment is equipped with robotic shelves mounted on servo-driven actuators and guide rails. These shelves enable the three-dimensional movement of drug containers and thus the selective removal and redistribution of inventory within the device. The compartments are also equipped with environmental sensors that continuously monitor parameters such as temperature, humidity, and vibration. These sensors are connected to an edge processing module embedded in the housing, which locally executes lightweight convolutional neural networks trained for anomaly detection.By running these models, the Edge module can detect cold chain breaks, temperature gradients, or vibration-related instabilities in near real time and initiate immediate corrective actions, such as activating thermoelectric cooling units or relocating inventory to compliant compartments.

[0049] The central processing unit includes a neural processing accelerator (GPU) configured to execute demand forecasting and optimization techniques. The forecasting engine is based on a recurrent neural network architecture and specifically leverages Long Short-Term Memory (LSTM) units to capture temporal dependencies in drug demand data. Historical prescription records, pharmacy sales data, distributor shipping logs, and epidemiological indicators are collected and processed to learn demand patterns over daily, weekly, and seasonal periods. The recurrent neural network is trained using backpropagation through time (BPTT) with gradient clipping to avoid instabilities and includes attention mechanisms that allow the system to prioritize sudden demand spikes due to disease outbreaks or vaccination campaigns.The output of the forecasting engine generates probabilistic demand distributions for each drug across different nodes of the supply chain.

[0050] The drive system consists of robotic storage racks driven by servo motors with integrated encoders, conveyor belts with load sensors and RFID tag readers, and deflection arms driven by stepper motors. When the central processing unit detects a redistribution, it sends pulse-width modulation signals to the servo motors, which in turn instruct the racks to retrieve specific containers. The conveyor belts are activated to transport the containers either to internal redistribution compartments or to external shipping modules, depending on the optimization decision. The RFID tag readers authenticate each storage unit against the blockchain-based traceability ledger, ensuring that only valid and compliant pharmaceuticals are redistributed. The load sensors confirm weight consistency, preventing errors such as underfilled or tampered packaging from being redistributed.

[0051] The environmental control subsystem embedded in the structural enclosure is integrated into the technical decision-making process. Thermoelectric Peltier cooling plates are distributed throughout the enclosure's insulated panels, along with closed air circulation ducts controlled by brushless DC fans. If the environmental sensors detect deviations, the edge processing module assesses the severity of the anomaly and either directly triggers a local action or forwards the response to the central processing unit for strategic decision-making. For example, if the temperature in one compartment rises above acceptable limits, the system can simultaneously activate the Peltier elements to restore conditions and trigger a redistribution command that moves critical vaccines to another compartment with stable thermal integrity.

[0052] The communication gateway acts as an interface between the device and external supply chain nodes. It is equipped with dual-band antennas that enable both 5G high-throughput communication and long-distance LPWAN transmission. The AI ​​engine uses the 5G channel to synchronize large data sets, including newly trained predictive models, with a central cloud repository. The LPWAN channel is used for continuous telemetry reporting to remote or rural pharmacies with limited bandwidth. The gateway features an adaptive routing module that automatically selects the most appropriate communication protocol based on available network conditions, ensuring uninterrupted operation in a wide range of environments.

[0053] The blockchain cryptography module embedded in the processing unit provides secure and immutable traceability. Every redistribution, whether internal or external to the device, is digitally signed using elliptic curve signature techniques executed on a dedicated cryptographic coprocessor. The digital signature is linked to environmental data—such as temperature and humidity at the time of relocation—and transmitted to the distributed ledger via the communications gateway. This ensures that regulators and stakeholders can audit every stage of the drug lifecycle, including inventory history, compliance records, and redistribution actions. The blockchain record not only prevents access to counterfeits but also strengthens confidence in automated decisions for inventory optimization.

[0054] The technical operation of the system follows a closed-loop process. First, data streams are collected from multiple sources, including environmental sensors, recipe systems, and merchant logs. These data streams are preprocessed at the edge to remove noise and detect anomalies. Second, the forecasting engine uses recurrent neural networks to generate demand forecasts for each product category at each supply chain node. Third, the reinforcement learning optimization engine evaluates redistribution policies in simulated environments, balancing competing objectives such as minimizing shortages while avoiding waste. Fourth, the central processing unit translates the selected policy into real-time control signals for actuators, conveyor belts, and environmental systems. Fifth, the cryptographic blockchain module immutably records each transaction.Finally, the system monitors feedback from both mechanical sensors and external demand signals to evaluate the success of the implemented policy and update the reinforcement learning parameters accordingly.

[0055] The edge processing module ensures that immediate responses to anomalies, such as cold chain breaches, can be executed autonomously and without cloud connectivity. At the same time, the central processing unit regularly uploads aggregated operational data to the cloud, where extended training pipelines retrain the recurrent neural network and reinforcement learning models using larger datasets. Updated models are then distributed to the devices via the communication gateway to ensure continuous adaptation to evolving global demand patterns. This hybrid framework enables the system to maintain its local autonomy in real time while benefiting from long-term improvements in prediction accuracy through centralized learning.

[0056] In one embodiment, the structural frame's modular storage compartments are dynamically reconfigurable. Linear actuators with sliding barriers divide the compartments into high-, medium-, and low-demand zones. The AI ​​engine continuously evaluates predicted consumption trends and adjusts the compartment boundaries accordingly. This adjusts storage capacity without the need to physically reassemble the unit. This adaptive compartment division ensures that high-demand products have more accessible storage space, while lower-demand products are consolidated, reducing unnecessary compartment cooling requirements.

[0057] By integrating advanced techniques with physical and environmental control, the invention transforms pharmaceutical inventory management into an intelligent, autonomous, and compliant process. The detailed architecture described above ensures that demand forecasts are not merely theoretical results but are translated into precise mechanical movements and environmental interventions. The system thus addresses the long-standing challenges of shortages, waste, compliance risks, and the threat of counterfeiting in pharmaceutical supply chains.

[0058] In one embodiment, the AI-based inventory optimization system is based on a hybrid cloud-edge architecture. Edge devices in warehouses and pharmacies are equipped with IoT gateways that can collect real-time environmental data such as temperature, humidity, and RFID-based item counts. These devices preprocess the data using lightweight convolutional neural networks (CNNs) for anomaly detection to identify early signs of cold chain violations or stock-outs. The preprocessed data is transferred to the cloud-based AI engine, which runs recurrent neural networks (RNNs) with attention layers to model demand fluctuations on a daily, weekly, and seasonal basis.

[0059] The AI ​​engine leverages multi-criteria optimization techniques to develop allocation strategies that minimize stockouts, reduce expiration losses, and ensure regulatory compliance. The optimization process is carried out using reinforcement-teaching agents that simulate various distribution policies and update strategies based on observed results. The system continuously learns from transaction and environmental data streams, improving its prediction accuracy over time.

[0060] The physical device consists of a structural frame configured as a modular storage and control unit. The housing integrates multiple compartments with robotic shelves that can automatically adjust storage placement. A series of belt-driven actuators, controlled by servomotor controllers, execute the storage relocation commands generated by the AI ​​module. The device also features thermal insulation and embedded Peltier-based cooling subsystems to ensure compliance with the cold chain in certain compartments.

[0061] The device's casing houses a GPU-accelerated central processing unit connected to a high-throughput communications module that supports both 5G and LPWAN connectivity for real-time cloud synchronization. A blockchain-based traceability registry integrated into the system ensures that every unit of medication passing through the device is digitally recorded and verifiable for regulatory inspections.

[0062] The technology controlling the device follows a closed-loop process: (1) environmental and transaction data collection, (2) preprocessing and harmonization of inputs from multiple sources, (3) demand forecasting based on deep learning, (4) optimization of distribution policies based on reinforcement learning, and (5) control of robotic shelves and conveyor belts to implement optimized allocation strategies. For example, if a sudden increase in antibiotic demand is forecast for a particular hospital pharmacy, the system triggers an actuator-controlled redistribution of medications from nodes with lower demand to nodes with expected high demand, thus maintaining balance throughout the supply chain.

[0063] In addition to redistribution, the system also enables predictive replenishment planning through its connection to upstream pharmaceutical manufacturers. As soon as the AI ​​engine predicts an impending shortage, it automatically generates replenishment orders and communicates them securely via the blockchain-enabled ordering module.

[0064] The drawings and the foregoing description illustrate examples of embodiments. Those skilled in the art will recognize that one or more of the described elements may well be combined to form a single functional element. Alternatively, certain elements may be separated into multiple functional elements. Elements of one embodiment may be added to another embodiment. For example, the order of the processes described herein may be changed and is not limited to the manner described herein. Furthermore, the actions of a flowchart need not be implemented in the order shown; nor do all actions need to be performed. Also, actions that are not dependent on other actions may be performed in parallel with the other actions. The scope of the embodiments is in no way limited by these specific examples.Numerous variations, whether explicitly stated in the specification or not, such as differences in structure, dimensions, and use of materials, are possible. The scope of the embodiments is at least as broad as indicated in the following claims.

[0065] Advantages, further benefits, and solutions to problems have been described above with respect to specific embodiments. However, the advantages, advantages, solutions to problems, and any components that may cause a particular advantage or solution to occur or become more apparent are not to be construed as critical, required, or essential features or components of any or all of the claims. REFERENCES 100 An AI-based inventory optimization system for pharmaceutical supply chains. 102 structural frames 102a Variety of modular storage compartments 102b Robot shelves 104 array of environmental sensors 104a Edge processing module 106 Central unit 108 Conveyor belt dispensing unit 110 Communication Gateway 112 Blockchain-integrated cryptographic module

Claims

[1] An AI-based inventory optimization system implemented as a device for pharmaceutical supply chains. The system includes: a structural frame defining a multitude of modular storage compartments, each compartment being equipped with robot racks mounted on motorized drive assemblies, the drive assemblies being configured to move drug containers along three-dimensionally guided rails; an arrangement of environmental sensors located in the compartments, wherein the sensors include temperature sensors, humidity sensors and vibration sensors, each sensor being electrically connected to an edge processing module housed in the frame, the edge processing module being configured to preprocess sensor data using convolutional neural network models to detect deviations in compliance with the cold chain; a central processing unit comprising at least one GPU or neural processing accelerator, wherein the central processing unit executes a demand forecasting engine based on recurrent neural networks and an optimization engine based on reinforcement learning agents, wherein the processing unit generates control signals for redistribution policies that minimize predicted bottlenecks, waste, and regulatory violations; a conveyor-driven dispensing unit operationally connected to the actuator units, wherein the dispensing unit comprises belt actuators, RFID tag readers, and load sensors and is configured to redistribute goods between compartments or to external dispensing points; a communications gateway that supports both 5G and LPWAN connectivity protocols for real-time synchronization with distributed supply chain nodes and cloud servers; and a cryptographic module embedded in the central processing unit and integrated into the blockchain, configured to digitally sign and record every inventory movement, environmental state, and redistribution command in a distributed ledger, thus ensuring immutable traceability of the drug inventory. [2] System according to claim 1, wherein the robot shelves are constructed as modular panels mounted along vertical and horizontal guide rails, wherein each shelf is driven by servomotors controlled via pulse width modulation signals, wherein the servomotors are also equipped with position encoders to provide real-time feedback to the central processing unit, and wherein the shelves are configured to selectively retrieve, relocate and redistribute drug containers under AI-generated control guidelines that adaptively redistribute inventory to compartments of high and low demand in response to predicted consumption trends. [3] System according to claim 1, wherein the environmental sensors are networked via a local bus architecture that connects each compartment to the edge processing module, wherein the edge processing module performs simple anomaly detection techniques that identify deviations in temperature gradients or sudden vibration events, and wherein the module is also configured to output corrective actuation signals that either activate Peltier-based thermoelectric cooling subsystems, initiate subdivided recirculating fans, or trigger the relocation of goods from an impaired compartment to an adjacent compliant compartment, thereby maintaining the integrity of the cold chain in real time. [4] System according to claim 1, wherein the conveyor-driven dispensing arrangement comprises a plurality of belt drives arranged in parallel channels, each drive being mechanically connected to a motor with torque control feedback, the belts being equipped with embedded RFID tag readers and load cell sensors configured to verify the authenticity and quantity of the transferred drug packages, and wherein the arrangement further comprises a mechanical deflection arm driven by a stepper motor, the deflection arm being controlled by the central processing unit to direct drug units either to shipping modules for external delivery or to redistribution channels for redistribution within the device. [5] System according to claim 1, wherein the cryptographic module integrated into the blockchain comprises a hardware-secured coprocessor implementing elliptic curve digital signature techniques, the coprocessor being configured to generate unique digital fingerprints for each drug relocation operation, the fingerprints being linked with environmental compliance metadata and being transmitted via the communication gateway to a distributed blockchain ledger, and the ledger ensuring that regulatory authorities and supply chain stakeholders can verify an immutable record of the storage history, compliance status and distribution route of each drug unit. [6] System according to claim 1, wherein the modular storage compartments are further subdivided into dynamically reconfigurable sections, each section being bounded by actuator-controlled sliding barriers driven by linear actuators guided on precision rails, and wherein the central processing unit actuates the barriers to change the size of the compartment volumes in response to predicted demand fluctuations, thereby enabling an adaptive division of storage capacity into high-demand, medium-demand and low-demand areas without requiring a physical reassembly of the device structure. [7] System according to claim 1, wherein the edge processing module and the central processing unit operate in a hybrid cloud-edge configuration, the edge processing module being configured to perform anomaly detection and corrective action in near real time in the presence of limited connectivity, while the central processing unit offloads aggregated data to cloud-based servers to regularly retrain the recurrent neural network predictive models, the updated models being redistributed to the device via the communication gateway, thereby ensuring continuous learning and adaptation without interrupting local operational autonomy. [8] System according to claim 1, wherein the structural frame of the device consists of a thermally insulated housing made of composite materials, wherein thermoelectric cooling plates based on Peltier elements and distributed heat sensors are embedded in the housing in each wall segment, and wherein a closed air circulation system comprising brushless DC fans and conduit channels is also integrated into the housing, wherein the circulation system is actuated by the central processing unit in response to sensor data to maintain a uniform temperature in all compartments, thereby ensuring full compliance with the requirements of the pharmaceutical cold chain and simultaneously supporting the mechanical relocation of stock.

Images