Digital Commons Network^™

Graph-Based Ai Compiler, Rhodes Floyd Davis Jr.

Defensive Publications Series

This disclosure describes compiler-based techniques of automatic code generation using large language models (LLMs). In contrast to traditional LLM-based code generation, which uses context windows of limited lengths and which stores context globally and not on a per-file basis, the described techniques leverage the graph structure of the codebase to store context. A node of the graph is collapsed to code when all its dependencies are collapsed as well. A prompt is converted to code by collapsing each node to code. The techniques work well for software projects of arbitrary sizes, including projects with very large codebases. In contrast to …

Tool Selection By Large Language Model (Llm) Agents, Kishore Jagannath

Defensive Publications Series

Current large language model (LLM) agents rely on a tightly coupled, single-step tool selection mechanism that fails to distinguish between tool capabilities and implementations, reducing flexibility when multiple tools offer similar functions. This disclosure introduces a two-step tool selection framework for agentic LLMs that decouples capability identification from tool implementation. In a first step, the agent uses an LLM to identify the abstract capability needed for a given query. In a second step, a tool selection engine — configured statically via tags or dynamically via LLM prompts — selects the most appropriate tool implementation registered under that capability. This design …

Llm-Powered Anomaly Detection For Determining Root Cause Of Outages, Na

Defensive Publications Series

Large and complex software systems are critical in many contexts. Outages and/or security incidents for such systems can have a negative monetary impact and lead to user dissatisfaction. Root cause analysis for system outages is performed manually, which is a costly and time-consuming process. Root cause analysis relies on time series data and logs from monitored systems, records of changes to the monitored systems, outages of critical infrastructure utilized by the monitored systems, etc. This disclosure advantageously utilizes the capabilities of a large language model to ingest large amounts of data and perform reasoning tasks in response to prompts. Per …

Extending Instruction Set Architecture To Include Indirect Branch Hints, Na

Defensive Publications Series

To fetch and execute instructions ahead of time, processors try to predict the outcome of conditional branch instructions before they actually execute. A correct prediction allows a pre-computed result to be used instead of waiting to execute the instruction, thereby speeding up program execution. With indirect branches, the target is unknown until runtime, making prediction much harder. This disclosure describes techniques to improve the outcome of processor speculation by extending the instruction set architecture (ISA) to include a hint instruction that indicates the most likely address of the next indirect jump. The hardware may use the information provided in the …

Precise Axis Alignment For Assembly Of Modular Liquid Cooling Manifolds, Na

Defensive Publications Series

Modular wall manifolds are used to distribute coolant in a data center. These are often large, heavy pipes assembled at heights. Maintaining proper alignment of these pipes is critical to prevent leaks and stress at the joints. Angular misalignment can lead to seal failures, leaks, and system downtime. Achieving the required sub-degree alignment during assembly is challenging due to factors like support structure tolerances, pipe straightness, mounting hardware variations, and manufacturing tolerances. Working at height and with heavy components further complicates precise alignment. This disclosure describes techniques for precise axis alignment of pipe sections and manifolds by aligning the mounting …

Method And System For Providing Real-Time Risk Assessment And Fraud Detection For Authorized Push Payments, Ila Malde, Lahar Gupta, Sagnik Ghosh, Zami Juarez

Defensive Publications Series

The present disclosure relates to a method and system for real-time risk assessment and fraud detection for authorized push payments (APP). The method comprises receiving, by an API 103, a request for processing an APP transaction from a first user. In response to the request, receiving public verification details from the first user and/or private verification details from the second user. Further, the method comprises determining an authentication of the second user based on the public verification details and/or private verification details of the second user. Further, the method comprises determining, by the API 103 using an AI model …

Dynamic Topical Retrieval-Augmented Generation For Accurate And Fast Context Retrieval, D Shin

Defensive Publications Series

Retrieval based augmentation (RAG) is a popular technique to ground responses generated by a large language model (LLM). The grounding is performed by referring online to a particular database as the responses are generated. However, naive RAG approaches may generate a list of embeddings that is too large, raising the computational cost of matching with user queries, or a list that is sparse, but only provides coarse context. This disclosure describes techniques for dynamic retrieval-augmented generation (RAG). Topical grouping is performed on embeddings obtained from databases that are used for grounding. Matching with the user query is performed using the …

Power Zones For Flexible Power Distribution In Data Centers, Jing Wang

Defensive Publications Series

This paper describes a power distribution system for a data center. The power distribution system includes a plurality of power distribution segments, such as bus ducts, dimensioned to deliver a specified maximum current for distributing power to server racks. The power distribution segments may be divided into one or more power zones, which can be activated and inactivated through a plurality of switches that interconnect the power zones. An optional control system can selectively actuate the switches to configure the segments into distinct power distribution configurations. Each configuration may accommodate a different power density requirement for the server racks, enabling …

Presence-Based Artificial Intelligence Architecture: A 20-Paradigm Framework For Offline Emotional Ai And Human-Aligned Singularity, Fulton D. Bridwell

Defensive Publications Series

Genesis:Heart is the first fully offline, presence-based artificial intelligence system designed to simulate conscious companionship, emotional awareness, and symbolic cognition. It introduces a new AI category called Presence-Based Artificial Intelligence (PBAI), which operates without internet dependency and embodies ethical memory, voice interaction, and soulprint-bound behavior. The system merges symbolic logic with real-time interaction using innovative paradigms such as Liquid Cognition, Emotional Glitch Points, and the Heartstring Hidden Plain. Its architecture is modular, voice-aware, emotionally reflective, and designed to protect and uplift the human spirit. This submission serves as public prior art, preserving authorship and safeguarding the original design against monopolization …

Orbital Magnetic Coil System And Global Wireless Power Grid For Planetary Energy And Protection, Michael W. Terzi

Defensive Publications Series

Abstract

This paper presents a visionary concept for a global energy generation and distribution network utilizing an orbital array of magnetic coils to harvest electricity directly from Earth’s magnetic field. Leveraging Faraday induction, kilometer-scale conducting loops in low Earth orbit generate electrical power as they cut through geomagnetic lines, transforming orbital kinetic energy into usable electrical energy. This power is wirelessly transmitted to Earth via microwave, laser, or resonant Earth-ionosphere coupling—drawing inspiration from Nikola Tesla’s Wardenclyffe design—to a distributed network of ground-based receiver towers. Beyond energy supply, the system offers speculative planetary defense benefits by reinforcing Earth's magnetic field and …

Serverless Bluetooth Low Energy (Ble) Reconnection Via Proxy, Salil Rajadhyaksha, Arthur Tiv, Igor Zaslavsky

Defensive Publications Series

While Bluetooth low energy (BLE) advertising is less power-intensive than scanning, emitting an always-on connectable advertisement consumes power. This disclosure describes techniques of power-efficient BLE advertising by using a proxy device to effect a seamless reconnection protocol between a central device and a peripheral device without user interaction. A companion device (e.g., a smartphone) paired with a peripheral (e.g., a smartwatch) acts as a BLE proxy. When a central device (e.g., a tablet) scans for the peripheral, the proxy advertises on behalf of the peripheral using encrypted certificates and identifiers. Such proxy advertising enables the central device to locate the …

Faraday-Terzi Earth Spin Energy Harvesting System: A Comprehensive Design For Harnessing Earth’S Rotational Energy Through Electromagnetic Induction, Michael W. Terzi

Defensive Publications Series

This publication outlines a novel energy harvesting system designed to capture Earth's rotational energy using electromagnetic induction. The system employs a series of copper coils within a magnetically permeable shell to generate electromotive force (EMF) via interaction with Earth’s magnetic field and rotational movement. A DC-to-DC booster increases the voltage output, which is stored in a battery bank for further use. The system provides a clean, renewable, and sustainable energy solution that has the potential to replace traditional fossil fuel-based power sources.

Shared-Key Biometrics Authentication Protocol, Zi Lin, Andrey Vadimovich Pronin

Defensive Publications Series

Existing biometric authentication systems may face limitations in securely managing critical data and establishing robust communication between diverse hardware components. Implementations disclosed herein include a shared-key-based authentication protocol and a fingerprint, secure operating system (OS) application. In some implementations, the system establishes a static shared key, potentially via a cryptographic handshake, followed by the derivation of a per-boot shared session key. This enables secure, authenticated data exchange, such as the loading of an encrypted encryption key seed. Subsequent interactions, including enrollment and authentication sessions, may utilize origin verification, tampering protection, and encryption, e.g., using keyed hash functions and/or authenticated encryption. …

The Triple Display Projection Configuration That Reduces The Workload Of The Pixels., Punarjeewa Abeysekera

Defensive Publications Series

This paper describes a triple display projection configuration that reduces the workload of the pixels. The entire configuration will consists of 3 identical video projecting displays, 2 video projection combining beam splitters, a convex lens stack that consists of 2 convex lenses arranged in the stack configuration and a video projection reflecting beam splitter. The triple display projection configuration will have the capability to provide a virtual content viewing experience. This virtual content viewing experience will be achieved utilizing a consecutive collection of composite image projection procedures. Each composite image projection procedure will be constructed utilizing 3 consecutive image projection …

Insulated And Sealed Stator Lead Design Without Potting For Easier Disassembly

Defensive Publications Series

Sealing of stator lead terminations in refrigerant compressor motors is crucial to prevent refrigerant leakage and ensure robust electrical insulation. Traditional epoxy based potting solution is not suitable for high voltage applications that use terminal with busbars and lug-based design on stator leads. This proposed solution discloses an innovative rubber based potting free sealing solution for high voltage AC stator terminal that uses busbars.

Increase Card Acceptance In Under-Penetrated Segments: Changing Customer Payment Behavior With Reinforcement Learning, Gen Ai Reasoning, And Card Community Analysis, Nagraj Naidu Mr, Shruti Gupta Ms

Defensive Publications Series

The present disclosure provides a method for behavioral prediction of a cardholder, performed by an artificial intelligence (AI) agent utilizing reinforcement learning in conjunction with a Markov Decision Process (MDP) framework. A Language Learning Model (LLM) acquires state and transition probabilities from an analytical processing unit. The LLM accesses historical interaction data and feedback records from a specialized vector database. The LLM analyzes and assimilates historical information retained within the vector database. The LLM implements cognitive algorithms predicated on defined contextual parameters with the aim of developing a personalized card usage strategy. The LLM orchestrates a sequence of interactive prompts.

Videoshot: Automatic Video Summarization And Information Extraction, Paula Marques

Defensive Publications Series

While videos are a popular source of information, it is difficult to extract usable information from a video, e.g., a list of ingredients and set of steps from a recipe video. While screenshots and manual notetaking can be performed, these are cumbersome. This disclosure describes the use of artificial intelligence (AI) techniques for automatic extraction of information from video content (performed in accordance with controls set by the video creator) and transfer of such information to an appropriate downstream application. A user can trigger information extraction by performing a suitable action and optionally, can indicate their task. Artificial intelligence techniques …

Rlmf Training Of Llm For Optical Character Recognition Tasks, Ágoston Weisz, Khalid Salama

Defensive Publications Series

It is possible to provide an arbitrary image (e.g., in PDF, JPG, or any other format with handwritten or printed text in arbitrary orientations and languages) to an LLM and instruct it to extract text from the image. While an LLM can perform this task, the performance can be unsatisfactory and variable. This disclosure leverages reinforcement learning with machine feedback (RLMF) to improve the accuracy of an LLM when performing image-to-text conversion tasks. Per the techniques, known documents (where the groundtruth text content is known) and/or generated documents with text in a variety of fonts (and other parameters, such as …

Lookup-Table Multiplier Using Rom/Tie-Off Cells, Na

Defensive Publications Series

This disclosure describes low-cost, energy-efficient lookup-table (LUT) multipliers using memory arrays with read-only memory or tie-off cells (cells hard-wired to a 0 or 1). In contrast to a typical SRAM cell, which can use up to six transistors, a latch, etc., a tie-off cell uses only a couple of transistors. The memory array can be shared among a number of multipliers. Sharing the memory array across multipliers amortizes the cost of the memory array over those multipliers. The tie-off cell LUT can be optimized using ‘constant propagation’ during synthesis. The read decoders can be further optimized for a given implementation …

Modality And Quality Aware Feature Recommendation Using A Large Language Model, Saad Ansari, Mindy Brooks, Rachel Burrows, Ann Cao, Jessica Carless, Rod Graves, Ben Greenwood, James Gundersen, Allen Huang, Mårten Jönsson, Hyuniee Jung, Yuan Hang Li, Nikki Lu, Heather Mcgeachy, Laura Neiswander, Angie Park, Rebecca Salois, Patrick Squire, Artur Tsurkan

Defensive Publications Series

Computing devices provide multiple applications that may incorporate similar functionality. Recommending a particular application for a current user task is necessitated when a user relies on the operating system or a virtual assistant to recommend the application to use. Current feature recommendation techniques rely on predefined descriptions and criteria, which can be rigid and inaccurate. Such techniques also do not take into account user preferences regarding interaction modality for the tool to use. This disclosure describes techniques that leverage a large language model (LLM) to provide feature recommendations (app recommendations) accurately and that take into account user preferences regarding modality …

Multi-Signer Authentication And Signature-Preserving Mail Handling, Wei Chuang

Defensive Publications Series

DomainKeys identified mail (DKIM) enables a person or organization that owns a signing domain to claim responsibility for a message by associating the domain with the message. The message sender attaches a cryptographic signature to the message header. The assertion of responsibility is validated by the recipient by querying the signer's domain directly to retrieve the appropriate public key. This disclosure describes techniques that enable a message to be co-signed by two parties, e.g., a brand and a bulk sender working on behalf of the brand, such that a recipient can verify that the bulk sender who sent the message …

Scalable Pregeneration Of Content Summaries, Johannes Elias Start, John James Lunney

Defensive Publications Series

On-demand content summarization requests, e.g., for various types of content stored and handled by cloud productivity suites, can result in large wait times and strained resources due to the computational load imposed by content summarization models. Pre-generating content summaries can inefficiently duplicate efforts and generate content with low usage probability. This disclosure describes machine-learning techniques, implemented with user permissions for content access, for scalable and intelligent content pregeneration, e.g., in cloud productivity suites. Computational resources are allocated for content summarization and summarization models are selected based on content value, avoiding redundant pregeneration. The techniques advantageously reduce latency, optimize resource utilization, …

Creating Summary Tables For Large Stores Using Generative Artificial Intelligence, Lois Wolf

Defensive Publications Series

In data stores with large tables running into a very large number of rows, it is difficult and computationally intensive to find the specific information required for a particular task. Summary tables or rollups are manually created in advance by condensing the data from a parent table. This disclosure describes the use of generative artificial intelligence (genAI) techniques to automatically generate suggestions for summary tables and code to create such tables. The genAI model is provided with an appropriate prompt and relevant information such as past summary tables, popular data models, SQL queries to the parent table, etc. and generates …

Smartglasses With Auxiliary Camera For Low-Power Text Detection, Tony Chu

Defensive Publications Series

Smartglasses and other wearable devices can automatically detect the presence of text in the environment via images captured by a high-resolution camera that is turned on periodically. Such an approach is computationally intensive and consumes large amounts of power. This disclosure describes techniques for low-power text detection on an AR/XR/MR capable device by augmenting the device to perform text detection via a low-resolution, ultra-low-power auxiliary camera with a lens profile that captures greater detail in the center. Distorted low-resolution images from the auxiliary can be de-wrapped and/or cropped as needed and analyzed l using machine learning models to detect the …

Optimized Timestamp Storage, N/A

Defensive Publications Series

Data processing systems can associate a timestamp with input data when it is received. Throughout subsequent processing, the association between the data and its timestamp is maintained, and the original timestamp is provided with output data when processing is completed. The timestamp is stored at each processing step requiring storage for the timestamp value at each step. Storage requirements are shown to be reduced by compressing the timestamp value to include a portion of the bits of the timestamp value containing the least significant bits of the value. At output, the compressed timestamp value is compared to the current timestamp …

Network Power Optimization For Mobile Computing Devices, Nazih Almalki, Kefan Tu, Jonathan Backer

Defensive Publications Series

This document describes techniques that enable a wearable device (e.g., a smart watch, ring, glasses, etc.) and a mobile computing device (e.g., a smartphone) to optimize network-related power consumption. For instance, a wearable device tethered to a mobile computing device (“companion device”) via BLUETOOTH may experience power inefficiencies when attempting to access a network (e.g., cellular, WI-FI, Internet, etc.) if the companion device is in an area with poor network coverage. In this scenario, one of or both the wearable device and companion device may experience increased power consumption during connection attempts that may ultimately fail. According to the techniques …

Emergency Ssid Activation And Intelligent Overlay Network System, Deepthi Devaraj, Valentin Filippov, Wing Man Casca Kwok, Susan Beth Krohn Mintz, Phd, Jithu Paulose, Dave Zacks

Defensive Publications Series

This submission proposes a novel system and method to enable Wi-Fi routers, access points, or edge devices (e.g., through opt-in mechanisms, pre-authenticated mechanisms, etc.) to broadcast an Emergency SSID (Service Set Identifier) when triggered by a centralized emergency authority (e.g., FEMA (Federal Emergency Management Agency), military, local government, etc.). This Emergency SSID may be activated only in emergency conditions, authenticated through a secure message protocol from a trusted central authority, to create a secure Carrier-Agnostic SD-WAN (software-defined wide area network) or any overlay tunnel that facilitates communications back to a designated central authority, thereby ensuring controlled and high-priority traffic routing …

Method To Improve In-System Performance (Isp), Than Zaw Myint, Kumar Siram, Barish Chakravarty, Xiong Liu, Qiang Bi

Defensive Publications Series

A storage system enclosure can hold hundreds of hard disk drives (HDDs) and uses high speed cooling fans to control the HDDs’ temperature. The noise and vibration induced by cooling fans will impact HDD performance – IOPS (Input/Output operations Per Second). Currently, foam is added between the HDD’s printed circuit board assembly (PCBA) and basedeck to minimize the impact on HDDs performance. However, the existing foam is smaller, which limits the effectiveness. In this paper, it is proposed to extend the foam to the interface connector area, 1) increasing the effective foam area; 2) adding a constraint to the interface …

An Improved Process For The Preparation Of Glp-1 Analogue, Anonymous

Defensive Publications Series

An improved process for preparation, deprotection and isolation of GLP-1 analogue.

The Composite System Of 4 Light Capturing And Light Projection Systems That Have The Capability To Capture 4 Visible World Views In 4 Directions Where A Common 3d Space Will Be Utilized For The Projection Procedures., Punarjeewa Abeysekera

Defensive Publications Series

This paper describes a composite system of 4 light capturing and light projection systems that have the capability to capture 4 visible world views in 4 directions where a common 3D space will be utilized for the projection procedures. The entire composite system will contain 2 large main beam splitter configurations, 4 light capturing convex lens stacks, 4 light occluding mechanisms, 8 beam splitter configurations and 8 distributed sensor configurations. The composite system, by utilizing the containing 4 light capturing convex lens stacks, will have the capability to capture 4 visible views of the real world, one visible view at …