Digital Commons Network^™

Llm Verifier Agent To Improve Business Attribute Extraction From Images, Na

Defensive Publications Series

Digital map providers enable entities such as businesses to add or edit their information. However, many businesses may not include their information in the digital map. This disclosure describes techniques that leverage a large language model (LLM) to improve text extraction from storefront images (e.g., taken from the street) and specifically, to extract business attributes from such images. Operating independently or supporting other text extraction procedures, the LLM is applied to verify text extracted from images. The described techniques can improve precision-recall in text-extraction problems.

Semantic Sensors: Multimodal Language Model Powered Sensors Capable Of Reasoning, Alex Olwal, Michael Xieyang Liu, Savvas Petridis, Vivian Tsai, Alexander Fiannaca, Michael Terry, Carrie Cai

Defensive Publications Series

While sensor-driven systems such as motion detection, home security systems, etc. are capable of preconfigured tasks, such systems offer limited customization. Large language models are capable of performing reasoning tasks based on data provided to them. This disclosure describes a comprehensive semantic sensing system that is configured to understand user intent and to perform reasoning tasks based on different streams of sensor data (e.g., obtained from low-level sensors) to fulfill that intent. Reasoning tasks can be performed using a foundational AI model. Semantic sensors - sensor-systems that utilize a multimodal language model that interprets sensor data - are described that …

Directional Flocking And Structural Integration Of Flocculated Polyacrylate Superabsorbent Fibers In Multilayer Absorbent Articles, Hannah Langford Dr

Defensive Publications Series

This study presents advanced multilayer absorbent articles utilizing flocculated polyacrylate superabsorbent fibers (SAFs), electrostatically flocked to adhesive surfaces or backing layers. SAFs were deposited on an adhesive on the non-contact side of foam layers or directly onto adhesive-coated backing films to optimize fluid uptake, retention, and structural integrity while protecting sensitive tissue from direct exposure to the absorbent material. A secondary design pathway includes the integration of flocculated SAFs within the foam matrix during manufacture to enable internal directional wicking. These layered systems are enhanced through silicone contact films, hydrocolloids, barrier films, and optional antimicrobial elements. Results demonstrate significantly improved …

Highly Efficient Mechanism For Cooling High-Power Pluggable Optical Transceivers Via Enhanced Thermal Contact At Elevated Ambient Temperatures, Anonymous

Defensive Publications Series

The present disclosure provides a highly efficient mechanism for cooling high-power pluggable optical transceivers operating at elevated ambient temperatures. Due to technological advancements, modern coherent optical transceivers consume significantly higher power, often up to 40W, while facing stricter junction temperature limits. Conventional cooling approaches, relying on enlarged heatsinks and increased fan speeds, have limitations, notably high thermal resistance at the heatsink-to-optics interface. The disclosed solution introduces a torsional spring-based lifting mechanism enabling controlled, high-pressure contact between the transceiver and a heatsink equipped with high-conductivity thermal interface materials (TIMs). Twin U-shaped heat pipes soldered into copper bases enhance uniform heat dissipation …

Xorcompactor: A Zero-Copy, Reversible Memory Defragmentation Technique For Operating Systems, Mostefa Sayeh Mr

Defensive Publications Series

This disclosure introduces XORCompactor, a novel, reversible, zero-copy memory defragmentation technique for operating systems. Unlike traditional memory movers such as memmove() or memcpy(), XORCompactor performs in-place compaction by leveraging a mathematical XOR-based transformation to swap and relocate memory segments without auxiliary buffers or double-copy overhead. This method ensures lossless reversibility, enabling safe rollback, debugging, and deterministic state recovery. The technique improves memory utilization, reduces fragmentation latency, and supports deterministic behavior in embedded and real-time systems.

XORCompactor includes the proposal of a new instruction-level operation, XOR_MOV, and outlines potential hardware integration for high-throughput memory management. The algorithm has …

Dlfp: A Deterministic Instruction Fetch Architecture For Safety-Critical Electronics, Mostefa Sayeh Mr

Defensive Publications Series

DLFP (Dual-Layer Fetch Pathway) is a novel deterministic instruction fetch architecture tailored for safety-critical systems requiring strict execution timing and resilience. It introduces the Instruction Heatmap Engine (IHE), a lightweight classification mechanism that segments instructions into Hot, Warm, and Cold zones based on execution behavior. By combining this classification with layered fetch queues (L0–L3), DLFP enables deterministic scheduling, thermal-aware promotion, and reduced fetch latency variability. This approach strengthens system predictability, mitigates speculative execution vulnerabilities such as Spectre and Meltdown, and enhances safety in real-time environments including SCADA, avionics, medical electronics, and autonomous robotics. DLFP offers a foundational architecture …

Solid State Forms Of (S)-(2-(5-Chloro-4-Methyl-1h-Benzo[D]Imidazol-2-Yl)-2-Methylpyrrolidin-1-Yl)(5-Methoxy-2-(2h-1,2,3-Triazol-2-Yl)Phenyl)Methanone Hydrochloride And Processes For Preparation Thereof

Defensive Publications Series

The present invention provides solid state forms of (S)-(2-(5-chloro-4-methyl-1H-benzo[d]imidazol-2-yl)-2-methylpyrrolidin-1-yl)(5-methoxy-2-(2H-1,2,3-triazol-2-yl)phenyl) methanone hydrochloride represented by the following structural formula-1a and processes for preparation thereof.

Efficient Inter-Switch Link (Isl) Congestion Notifications In Fibre Channel Fabrics, Venkat Garigipati, Harsha Bharadwaj, Srinivas Pitta

Defensive Publications Series

Handling congestion on an inter-switch link (ISL) port in a fibre channel (FC) fabric is not specifically addressed in current FC standards. Further, strictly following current standards can result in sending Fabric Performance Impact Notification (FPIN) signaling to endpoints that may not be contributing to ISL congestion. In order to address such issues, techniques are proposed herein that provide for efficiently handling ISL congestion using storage area network (SAN) analytics to accurately determine the flows that are causing ISL congestion and generating FPIN-Congestion Notification (FPIN-CN) messages that can be sent to the endpoints originating the flows causing congestion. The FPIN-CN …

Synthetic Data Generation For Training Video Understanding Models, Ivor Rendulic, Manuel Tragut, Ágoston Weisz

Defensive Publications Series

Obtaining training data to train video understanding models traditionally involves human experts that generate text based on a video. This is expensive, prone to human errors, and not easily scalable. This disclosure describes techniques that leverage machine learning models to automatically generate training data for training models for video understanding. Video and image captioning models and speech-to-text techniques are used to generate text captions for training videos. A large language model LLM) is prompted to generate a diverse set of questions based on the captions as well as questions unrelated to the video. An LLM grounded on the captions is …

High Aspect-Ratio Filler-Containing Conductive Elastomeric Materials And Methods Of Use

Defensive Publications Series

Described herein are conductive elastomeric materials and methods of using the same. The conductive elastomeric materials include a polymeric matrix, a peroxide accelerator, and one or more high aspect-ratio fillers, wherein the high aspect-ratio fillers are present in an amount of 50 wt. % or less based on the weight of the conductive elastomeric material. The polymeric matrix can include a thermoset polymer or a thermoplastic polymer. Also described herein are methods of making conductive elastomeric materials. Further described herein are molded products including the conductive elastomeric materials as described herein and wearable devices including the molded products.

Smart Conference Room Lighting For Enhanced Video Calls, Hp Inc

Defensive Publications Series

This disclosure relates to where the field of Video Conferencing and Smart lighting intersect. A program (algorithm) is disclosed that analyzes and processes information of conference call participants and integrates video camera software with smart home/office software. This enables the new software to provide superior conference experience by providing better lighting and presentation of participants than any of the existing alternatives.

Wheel Trim Removal Tool Integrated In Lug Nut Wrench Tool, Anonymous

Defensive Publications Series

This innovation project is looking to establish a permanent integration of a multi-functional tip directly into the wrench tool. This integrated tip offers adaptability to various wheel trim attachment designs as a universal tool. The integrated tool is pivotally connected to the wrench via a fastener, enabling rotation about its axis. This pivotal connection allows selective positioning of the tip in either a 'working' configuration or a 'storage' configuration.

Artificial Intelligence To Reduce Printer Ink Usage, Kevin Ngo, Ruifeng Xu, Benjamin M. Gordon

Defensive Publications Series

Print jobs, especially those involving printing images or non-standard text documents, can consume large amounts of ink resulting in a substantial expense that the user may be unaware of. This disclosure describes artificial intelligence (AI) techniques to optimize printer ink usage. An AI model is trained using a dataset of various types of print jobs, printers, and printer settings, and corresponding ink usage patterns. When a user sends a print job to the printer, the trained AI model is utilized to generate a prediction of ink usage. Depending on the type of print job, the AI model also generates alternatives …

Protecting Against Use-After-Free Exploits With Multiple Policy-Limited Memory Heaps, Joshua Mason

Defensive Publications Series

Bugs or errors in handling memory allocation and deallocation can lead to use-after-free (UAF) vulnerabilities. Such cases involve programs accessing memory locations that have been freed by deallocation. Malicious actors try to exploit UAF vulnerabilities to cause various adverse effects with potentially severe operational consequences. This disclosure presents an alternative protection approach by replacing the use of a single memory heap with multiple policy-limited memory heaps. At runtime, each of the distinct memory heaps can be randomly assigned a policy to limit the type(s) of objects that can exist within it. The random policy assignment reduces the likelihood of malicious …

Voice-Enabled Stylus For Contextual Command Input And Dynamic Content Enrichment, Hadi Zadeh

Defensive Publications Series

Current stylus devices lack integrated voice command capabilities, limiting the ability of users to efficiently interact with touchscreen devices while performing creative tasks with the stylus. This disclosure describes a voice-enabled stylus equipped with a microphone and a button for activating voice commands. Users can control the stylus by directly voicing commands to it. Unlike a general-purpose microphone on a mobile device, the stylus microphone supports specific stylus-related actions such as changing drawing tools, activating features like rulers, adding transcribed content during note-taking, etc. Commands can also be used to enrich diagrams or drawings with audio recordings, such as adding …

Enhanced Denoising With Voice Activity Detection, Sinan Hersek, Thomás Inskip

Defensive Publications Series

This disclosure describes techniques to enhance multi-channel speech denoising using voice activity detection (VAD). Per the techniques, VAD is leveraged to detect segments of microphone signals where the user’s speech is absent but where background noises are present. For such segments, the microphone output to the headset is set to silent. Effectively, VAD is used to gate the microphone output based on the presence or absence of the user’s speech. The gating action of the VAD results in a denoising, since noise is completely suppressed at least in segments where the user’s speech is absent. Denoising using VAD is described …

Audio Generation Using Piezoelectric Elements In Smartglasses, Craig Ranta

Defensive Publications Series

Integrating speakers into small-size devices is difficult. This is especially challenging for smart glasses, augmented/virtual reality (AR/VR) headsets, or similar devices due to the compactness of the arm or temple portion of the glasses or headsets. This disclosure describes techniques that leverage the piezoelectric effect to generate high-quality audio using components that can fit into the compact form factor of smart glasses. Smart glasses and VR headsets have glass or plastic lenses, which advantageously present an ample area for piezoelectric elements (crystals) to be affixed upon. When electrically excited, the piezoelectric elements cause the lenses to imperceptibly vibrate in synchronism …

Large Language Model Based Automated Labeling Of Keypoints In Images Of Rooms, Adam Zu, Fengtao Fan

Defensive Publications Series

Spatial applications require accurate geometrical measurements of physical spaces. Typically, room geometry can be determined based on the position of various corners in the room. A room keypoints model can provide the locations of corners in an image of a room. However, there is a dearth of diverse types of suitable images with labeled keypoints that can be employed to train keypoint models. Manual labeling does not scale because it is tedious, slow, and expensive. This disclosure describes an LLM-based agent to automate the labeling of keypoints, such as corners within room images at scale, thus enabling speedy generation of …

Cross-Realm Interaction Between Digital Wallet And Shopping Scenarios In Extended Reality, Yanhe Chen, Ruofei Du

Defensive Publications Series

While digital payment via mobile devices is common, there are no intuitive digital-payment mechanisms for new device categories such as augmented/extended reality (AR/XR) devices, smart glasses, etc. This disclosure describes cross-platform payment techniques designed for immersive extended reality (XR) environments. Per the techniques, a user can invoke a panoramic view of their linked credit cards within their XR display. The user can select their preferred payment method/instrument using intuitive gestures based on hand, eye, or head movements. The user can make a purchase from a virtual marketplace without interrupting the immersive experience. The XR headset can be linked to a …

Test Fixture And Test Methodology For Thin Glass Characterization, Maxim Sivov, Venkata Mk Akula, Benjamin H. Golub

Defensive Publications Series

This disclosure describes a test methodology and test fixture suitable for glass characterization of relatively thin glass materials. The test fixture for specimens measuring 0.5-0.7 mm in thickness includes an inner ring and outer ring with spans of about 5 mm and about 10 mm, respectively. For specimens measuring 0.3-0.4 mm in thickness, the inner ring span is about 4 mm and the outer ring span is about 8 mm. The fixture has independent degrees of freedom that allows the top ring carrying the loading force to compensate for any flexural deflection and loading misalignment. Additionally, the fixture enables precise …

Single Sideband Coherent Optical Detection, Na

Defensive Publications Series

This disclosure describes a coherent optical detection technique based on single sideband (SSB) reception that enables finer per-wavelength speed granularity while maintaining high efficiency in terms of power and components-count. By recovering real-valued pulse amplitude modulation (PAM) signals from either the lower or upper optical sideband, the SSB coherent receiver architecture reduces required component bandwidth and can achieve signal-to-noise ratio (SNR) performance comparable to or better than conventional coherent detectors. The technique supports integration with polarization-folding, time-diversity coherent optics to further enhance system scalability and reduce receiver complexity. This approach enables the design of cost-effective, high-speed, low-power transceivers for short-reach …

Managing Flash-Memory Reliability By Proactive Monitoring Of Bit Error Rate, Na

Defensive Publications Series

The lifespan of NAND-based flash memories is limited by physical properties of flash cells. The reliability of retrieving data stored in flash memories can deteriorate over time. This disclosure describes techniques to manage flash-memory reliability by monitoring bit error rate (BER) per physical memory block and by taking certain actions based on the observed BER. Such actions include ignoring BERs that are low enough to be recovered by an error-correcting code (ECC) without impacting performance or reliability and notifying the processor if the BER rises to high levels. Ignoring low BERs, e.g., not escalating low BER rates to the processor, …

Hybrid Reasoning System For A Large Language Model Operating System Incorporating Symbolic Inference And Knowledge Graph Integration, Krishna Chytanya Ayyagari

Defensive Publications Series

This disclosure describes an Adaptive Agent Runtime Environment (AARE) architecture. The AARE architecture (AARE) addresses several critical limitations and unmet needs in the burgeoning field of Large Language Model (LLM) integration into complex computational systems. AARE provides a comprehensive synthesis of various current technologies as well new concepts, including integration of various technologies. It seeks to overcome the limitations of existing approaches by offering a tightly integrated framework that natively supports autonomous agents, leverages the strengths of LLMs and symbolic reasoning, provides robust memory management, fosters extensibility through a plugin ecosystem, ensures security and isolation, and facilitates discovery and distribution …

Augmented Reality Platform To Improve Last-Meter Arrival Using Visual Navigation, Jaydeep Untwal, Rakhi Parekh, Yash Dave, Aakanksha Parameshwar

Defensive Publications Series

This document describes an augmented reality (AR) platform designed to enhance last-meter arrival experiences through visual navigation. The platform leverages computer vision techniques and mobile device capabilities to provide users with real-time guidance in complex or unfamiliar environments. The AR system analyzes the user's surroundings, performs a visual frame matching, and overlays contextual information onto the user's view, thereby facilitating seamless navigation. The platform includes components such as a visual tracking module, a route planning algorithm, and an AR rendering engine. By utilizing the system, users may experience improvements in wayfinding efficiency and reduced cognitive load during the final stages …

System, Method, And Computer Program Product For Evaluating Generative Machine Learning Models, Yang Wang, Alberto Garcia Hernandez, Roman Kyslyi, Nicholas Kersting, Ajit Vilasrao Patil, Ranjan Dutta

Defensive Publications Series

Systems, methods, and computer program products are provided for evaluating generative machine learning models. An example system includes a processor configured to determine queries and ground-truth answers associated with the queries. The processor is also configured to generate generated answers based on the queries using a model that is being evaluated. The processor is further configured to input the queries, the ground-truth answers, and the generated answers to an evaluator model trained to evaluate the generated answers in comparison to the ground-truth answers and based on a grading scale associated with accuracy, honesty, and completeness of a generated answer. The …

Empathy Tags For Granular Access Control For Collaborative Digital Objects, Dhandapani Shanmugam, Soundararajan Ramalingam

Defensive Publications Series

Collaborative editing of digital objects such as documents, slide presentations, etc. is common. Large, complex documents may have a large number of collaborators. Current document sharing technologies enable users to add collaborators with different levels of permission for the entire digital object. Per techniques of this disclosure, document metadata are updated to include respective empathy tags associated with different portions of the document. Each tag identifies specific users (or groups) and associated permissions for that portion. In this manner, granular content access control is enabled, allowing users fine-grained control over which collaborators have what level of access to different parts …

Enhanced Absorptive Performance And Conformability Of Flocculated Superabsorbent Materials In Personal And Medical Care Articles, Hannah Langford Dr

Defensive Publications Series

Incontinence, personal care, and wound dressing products require high absorbency and flexibility for optimal performance. This study investigates a novel approach to constructing absorbent articles using flocculated superabsorbent materials (FSAMs), comprising microparticles, granules, or molecular-scale polymers. These FSAMs are integrated into articles in distinct structural configurations, either as discrete layers, embedded within foam, or localized within foam channels or fenestrations. The combination of flocculation and integration within or adjacent to foam matrices significantly improves conformability, absorption kinetics, and total fluid retention compared to traditional nonwoven-based superabsorbent materials. Additional layers—such as silicones, hydrocolloids, films, and hydrogels—enhance the functional performance of the …

Battery Energy Storage System With Water Submersion Safety Design, Na

Defensive Publications Series

Lithium-ion batteries can have catastrophic impacts when they enter a thermal runaway event. Currently, when a thermal runaway event occurs, heat is removed from the chemical reaction by quenching the batteries with water from fire sprinklers in the facility whether the batteries are deployed. However, this can cause flooding of the entire building or facility in which the batteries are deployed, e.g., a datacenter that houses expensive servers and other equipment. This disclosure describes a battery energy storage system with a water-submersion safety design. An uninterruptible power supply with battery modules is installed above a quench tank filled with water. …

Conversational Incident Reporting Using A Large Language Model, Chaim Zucker, Nitzan Jaitman, Alon Segal

Defensive Publications Series

Some digital map applications include road incident reporting features that enable users to report incidents along a path that a user is traveling on. Such features require specific phrasing or manual selection of options on a device screen which can be cumbersome and distracting. This disclosure describes techniques that enable intuitive and safe road-incident reporting using hands-free, free-form natural language conversations within a digital map application. A reporting button on the user interface (or spoken command) of a digital map application enables users to start the reporting of incidents. A large language model is deployed to interpret the spoken incident …

Adaptive Decision Engine Using Large Language Models And Configurable Schema, Gunalan S, Akshatha B

Defensive Publications Series

This disclosure describes a framework to automate complex decisions by combining the strengths of human-defined logic with insights derived from large language models (LLMs). Human decision-makers define templates, known as ‘decision models’ or ‘decision schema,’ for the structure and the core logic of the decision-making process. The templates are augmented or populated using output derived from LLM analysis. The LLM can be used to provide insights, generate text, or fill in specific parts of a pre-defined template, thereby enhancing the human-designed logic. The framework uses the completed, LLM-enhanced templates to automate decisions efficiently and transparently. Effectively, LLMs augment and structure …