A Multi-Agent Framework For Dynamic Geospatial Acoustic Characterization, Lenord Melvix

Defensive Publications Series

Characterizing the acoustic environment of a geospatial region may be challenging due to limitations in some existing data sources, which can be outdated, geographically broad, subjective, or anecdotal. This document describes a multi-agent computational framework for dynamic geospatial acoustic characterization. The system can utilize a collection of specialized software agents to ingest, process, and synthesize data from multiple heterogeneous sources, such as street-level and aerial audio recordings, user-generated content, and official data feeds. For example, agents can perform tasks such as data sourcing, denoising, source-specific characterization (e.g., vehicular, industrial, human), and temporal synthesis. This framework can be used to produce …

System For Generating Themed Content Using A Generator-Evaluator Framework, Roy Katz

Defensive Publications Series

Manual creation of themed navigational content for user engagement campaigns can be a slow, resource-intensive, and unscalable process, while general-purpose generative models may lack domain-specific safety guidelines and struggle with thematic consistency. A disclosed system can address these challenges through a dual-model, generator-evaluator framework. A primary generative model can create multiple thematic text candidates for a given instruction. A separate, secondary evaluator model may then assess these candidates using a hybrid evaluation process that combines model-based semantic analysis for attributes, such as safety and logical contradiction, with function-based checks for criteria like length and semantic similarity. The system can iteratively …

Estimating Geospatial Entrances Via Multi-Agent 3d Contextual Alignment, Lenord Melvix

Defensive Publications Series

To address imprecisely placed entrance markers for geospatial entities in digital maps, for instance, for new or low-traffic locations with sparse data, a multi-agent computational system can be utilized. The system can process various data sources, including user-supplied media such as photos and videos, to generate a three-dimensional (3D) model of an entity's structure and identify potential entrances. Concurrently, the system may construct a 3D contextual model of the surrounding region using data such as building footprints and road networks. An iterative alignment process can position the entity's 3D model within the regional model based on a set of logical …

Ai-Powered Framework For Test Redundancy Detection And Optimization In Software Projects, Hp Inc

Defensive Publications Series

This disclosure presents an AI-powered solution that leverages Large Language Models (LLMs) and machine learning techniques to identify, analyze, and consolidate redundant test cases. In enterprise projects that are maintained for long period of time, test repositories accumulate significant duplication and redundancy, due to requirement changes in the project, which create significant operational challenges. The approach streamlines test repositories, optimizes tests, and reduces execution overhead. This methodology demonstrates improvements in long-running system tests where traditional manual optimization approaches become impractical due to scale and complexity.

System For Autonomous Generation And Methodologically Guided Optimization Of Language Model Prompts, Hui Wu, Yulou Zhou, Simon Zhao

Defensive Publications Series

The efficacy of large language models may be dependent on prompt quality, and manual prompt engineering can be an inefficient, trial-and-error process. Some existing automated optimization approaches may utilize a pre-existing seed prompt and could lack methodological guidance, potentially limiting their effectiveness. A system is described for the automated optimization of instructional prompts that can employ a closed-loop algorithmic pipeline to iteratively generate, evaluate, and refine candidate prompts. The process can be initiated, for example, without a human-provided seed prompt, instead originating initial candidates from reference data and a structured methodological framework, such as a chain-of-thought workflow. This automated and …

Automated Analysis Of User Logs Using Large Language Models In An Agentic Workflow, Sha Li, Chenjie Yu

Defensive Publications Series

Analysis of user interaction logs can be challenging because manual review may not be scalable and aggregated data analysis can obscure the chronological context of user behavior. Systems and methods for automated analysis can use a structured agentic workflow. This workflow can programmatically extract raw, time-ordered session logs and can utilize a large language model, guided by a prompt engineering subsystem, to perform analytical tasks such as classification, summarization, and reasoning on the sequential data. This approach can facilitate scalable and standardized analysis of user behavior by preserving the temporal detail of individual sessions, which may enable AI agent for …

Imu-Driven Watch Face With Performance-Optimized Graphics, Gabriela Namie, Teddy Guerrero, Seth Benson, Courtney Cox, Tyler Gough, Mike Humphrey

Defensive Publications Series

A smartwatch interface displays dynamic, three-dimensional-like animations which are directly responsive to the user's wrist movements. The system uses data from the watch's Inertial Measurement Unit (IMU) to drive the animation of on-screen graphical elements in real-time. The method includes code-based techniques for simulating visual effects like gradients and for managing element animation to maintain fluid performance (e.g., 30 frames per second) to overcome hardware limitations related to rendering complex graphics and a high number of elements.

Trust Based Intelligent Bidding Mechanism For Selecting Efficient, Highly Rated And Suitable Worker Agents, Aditya Kesarwani, Sushanth Patil, Sunil Pareek

Defensive Publications Series

A system including a trust-based intelligent bidding mechanism is proposed herein that addresses the critical challenge of selecting an appropriate agent within an agentic artificial intelligence ecosystem, where users and customers often struggle to evaluate agents, resulting in suboptimal outcomes. This innovative system incorporates bidding agents and a trust layer that facilitates the selection of trustworthy agents for user task execution. Thus, the system rapidly and accurately provides users with appropriate worker agents while considering other factors such as cost, ratings, and user feedback.

Unified Multi-Modal Adaptive Caregiving Assistance Platform, Kshitij Suresh Gaikar

Defensive Publications Series

The present disclosure relates to a unified, multi-modal, adaptive platform configured to monitor and analyze the needs of human infants and domestic pets. The system utilizes a sensing module to capture audio, video, and environmental data, which is processed through a hybrid edge and cloud computing architecture. By employing deep multi-modal data fusion, the system integrates synchronized feature vectors to infer specific subject needs. The platform generates ranked, actionable recommendations for caregivers and incorporates a reinforcement learning feedback loop to personalize suggestions based on subject-specific responses.

Keywords: Caregiving, Multi-modal Data Fusion, Artificial Intelligence, Reinforcement Learning, Infant Monitoring, Pet Monitoring.

Generative Model-Assisted Generation Of Structural Selectors For Web Data Extraction, Jordan Janeiro

Defensive Publications Series

The automated web data extraction can be challenged by time-consuming manual methods and direct data retrieval by generative models that may be unreliable due to outdated training data. A disclosed technique can address these challenges by using a generative model to analyze the document object model of web pages. Instead of extracting data content, the model can identify stable structural patterns containing desired data and can output corresponding selectors, such as cascading style sheets selectors or XPath expressions. This approach, which may involve clustering structurally similar pages before analysis, can separate structural pattern recognition from final value extraction. This process …

Account Risk-Based Watermarking Of Generative Ai Content, Pingping Chen, Meng Ya, Tyler Kress, Jonathan Joa, Chris Xu

Defensive Publications Series

Systems and methods are described that may link a user account's risk profile to the content it generates. The technology can assess account behavior to generate a risk score and can embed corresponding digital watermarks into AI-generated content. For example, an origin watermark may identify the content as AI-generated, while a safety watermark may encode information related to potential risks based on the account's score and the user's prompt. This approach may allow downstream platforms to detect these watermarks, which can enable them to apply context-aware moderation, such as displaying informational labels or providing user-configurable controls, facilitating safety enforcement at …

System For Autonomous Behavioral Verification And Remediation Of Migrated Database Code, Khushmeet Rekhi

Defensive Publications Series

The migration of database code may result in data corruption when static translation tools do not fully account for behavioral differences between source and target platforms, for example, in rounding or null value handling. A dynamic verification approach can address this. A system may passively observe a legacy database to capture truth vectors, which can be records linking transaction inputs to their verified outputs. Translated code can then be executed in an isolated sandbox environment using these inputs. A semantic comparator may analyze the new output against the legacy output from the truth vector. If a material discrepancy is found, …

Non-Invasive Dielectric Fluid Health Monitoring System, Dhruv S Chaddah, Zaid A Issa, Gowrav Bukkapattana Gururaj, Yaagna Krupal Modi

Defensive Publications Series

Proposed herein is a system that integrates high-resolution capacitive, optical, and infrared temperature sensing into a compact, clamp-on module that mounts directly onto a non-metallic coolant tube. Ideally, the clamp-on module can be located at the pump outlet, where fluid temperature, flow, and dielectric properties are most representative of system health. This fully non-invasive approach enables continuous, real-time monitoring without disrupting the cooling loop.

An Improved Process For The Preparation Of Momelotinib Dihydrochloride Monohydrate, Anonymous

Defensive Publications Series

The present publication relates to a process for the preparation of Momelotinib dihydrochloride monohydrate, chemically known as N-(cyanomethyl)-4-{2-[4-(morpholin-4-yl) anilino] pyrimidin-4-yl} benzamide dihydrochloride monohydrate (Formula I).

Conscious Ai Thought Origination In Llm, Shreyas Kc, Yu Gu

Defensive Publications Series

The present disclosure relates to the field of Artificial Intelligence (AI), and more specifically to conscious AI systems integrated with Large Language Models (LLMs) for autonomous ideation and decision-making. The present disclosure introduces a consciousness layer to a fine-tuned LLM, enabling the model to continuously generate candidate topics, rank them using a weighted scoring algorithm, and select the highest-priority topic for ideation. The consciousness layer captures and stores intermediate outcomes, embeddings, agent actions, and metadata, which are vectorized and re-injected into the LLM’s context to maintain continuity. This loop operates without external prompting, emulating humans like spontaneous thought and topic …

Automated Optimization Of Structured Large Language Model Prompts Via Iterative Evaluation, Hui Wu, Haoda Huang, Samuel Shen, Enrica Filippi, Imed Zitouni, Simon Zhao, Zach Fisher

Defensive Publications Series

Performance of generative models (GMs), such as Large Language Models (LLMs), is highly dependent on the quality of the input prompt that is processed using the GM. However, manual prompt engineering is often inefficient and requires specialized expertise. This disclosure describes a method for the automated optimization of structured prompts through an iterative evaluation process. A structured prompt template is initially populated using task descriptions and reference data to generate candidate seed prompts. These candidate seed prompts are executed against a targeted model, and the resulting responses are assessed by an automated evaluator. Based on performance metric(s), a refinement loop …

Reducing Stamping Friction Coefficient Via Optimized Grinding Approach, Fei Fan, Yan Lin

Defensive Publications Series

The present disclosure is related to a process for optimizing grinding operations that reduce the overall coefficient of friction and prevent cracking. It was found that changing the grinding direction to be perpendicular to the local sheet metal flow significantly reduces the coefficient of friction and helps prevent stamping cracks. For example, optimizing the grinding direction can help entrap lubricant, which has an observed effect on friction. In some critical areas, grinding along the feed direction can reduce wrinkling.

Rme-1: A Formal Specification For Geometric Semantic Refinement, Samuel J. Church

Defensive Publications Series

RME-1: A Formal Specification for Geometric Semantic Refinement

Document ID: RME-1-2026-PRIOR-ART-01

Status: Public Disclosure for Defensive Publication

Core Methodology: Non-Euclidean Mapping of Abstract Logic to Sensory-Grounded Vectors

I. Mathematical Framework: The RME Hilbert Space

The RME-1 engine operates by projecting linguistic units into a multi-axial Hilbert Space. Unlike standard token-probability models, RME-1 enforces a Geometric Constraint (GC) on the output, defined by the relationship between four primary variables:

1. Symbolic Density ($SD$): The concentration of domain-specific semantic tokens per unit of syntax.

2. Sensory Gravity ($SG$): The coefficient of tactile or physical-structural grounding in the output string.

3. Intent …

   ---

   Go to article

Quantifying User Journey Progression Using A Language Model, Lijun Wang, Qing Zhang, Thomas Vaughan, Kangxing Fu, Shuo-Heng Chung, Evgeny Skvortsov, Harikesh Nair, Shawn Xu, Jianqiao Wang, Jing Wang, Charles Huyi, Junyi Zhang, David Quintero, Jeff Yam, Jiro Root, Tong Geng

Defensive Publications Series

Systems and methods can quantify user journey progression in digital advertising analytics, as some static, rule-based models may not capture the non-linear nature of certain user behaviors. The disclosed technology can utilize a data-driven framework with a sequential prediction model, such as an n-gram model, to analyze user interaction data. Raw user actions can be ingested and mapped to discrete symbols, or grams, which can then be formed into chronological sequences representing user journeys. The model can process these advertiser-specific sequences to compute a transition probability table that quantifies the likelihood of a user's next action based on their recent …

Embedded Liquid Cooling Architectures For Printed Circuit Boards, N/A

Defensive Publications Series

This paper describes thermal management solutions for electronic devices, specifically focusing on liquid cooling mechanisms integrated directly into printed circuit boards (PCBs). The configurations detailed herein address thermal challenges by embedding cooling conduits and heat exchange structures within the PCB substrate itself. Such approaches may include embedding copper tubes inside copper coins in the PCB or may include a cold plate having inlet and outlet tubes installed into a cavity within the PCB.

Generation Of Synthetic Photographs Via Video Synthesis And Heuristic Frame Selection, Mira Leung

Defensive Publications Series

Generating novel synthetic photographs of locations can be challenging, as some generative techniques may introduce visual artifacts or compositional inconsistencies when creating new camera perspectives. The described systems and methods can address this by using two or more source images as input for a generative video synthesis model. The model can produce a short video that interpolates or extrapolates between the viewpoints of the source images, creating a set of new, spatially coherent camera angles. This generated video may then be deconstructed into a pool of candidate frames. A heuristic, such as a feature similarity score, can be used to …

Real-Time Intrinsic State Analysis Of Neural Networks Via Instrumented Characteristic Regions, Johannes Start, John Lunney

Defensive Publications Series

This document describes systems and methods for analyzing the internal state of neural networks, such as large language models, during production-level inference. Some existing analysis techniques may rely on extrinsic, black-box analysis of model outputs, which can be slow, costly, and may not detect certain performance degradations. The described technology can involve an offline process to identify and calibrate sparse characteristic regions within a model's architecture, where specific activation patterns may correlate with high-level internal states. During runtime, lightweight counters can be instrumented into the model's computational graph to measure activations within these regions, potentially with low performance overhead. This …

Device For Taking Images Of An Aircraft Engine, Julian Radziwill

Defensive Publications Series

A device for taking images of an aircraft engine, comprising a. a frame (1), in particular a mobile frame (2); b. a rotatable beam (2), in particular rotatably mounted to the frame, comprising i. a plurality of attachment elements, each attachment element configured to allow for a camera module (3) to be releasably mounted thereto.

Predictive Cost-Benefit Routing For Multi-Tier Risk Decisioning Systems, Subhadip Mitra, Kerrie Hudson, Archanaa Ravikumar

Defensive Publications Series

The real-time risk decisioning systems can encounter a trade-off between the high cost and latency of certain accurate analytical systems and the speed of simpler, deterministic rules, as some escalation methods may not account for the financial context of individual transactions. A disclosed technology provides a multi-tier risk decisioning system managed by a predictive, cost-benefit routing engine. This engine can calculate an expected return on investment for a transaction before escalating it to a more computationally expensive analytical tier. Escalation may proceed if the potential financial benefit of a more accurate decision is determined to outweigh the operational cost of …

Automated Rule Generation For Tiered Systems Using Multi-Stage Failure Learning, Subhadip Mitra

Defensive Publications Series

In multi-tiered anomaly detection systems, a challenge may exist in balancing the operational cost of accurate analytical models with the accuracy of less expensive rule engines, where manual rule creation can be slow. A described technology can address this by, for example, monitoring a multi-stage processing cascade and employing multi-stage failure learning. A system can identify instances where a low-cost analysis tier does not detect an anomaly that a higher-cost tier subsequently identifies. The system may then extract the successful detection pattern and use a generative model to formulate a new, low-cost rule that codifies this logic. This process can …

Method And System For Secure Re-Registration Of Unified Payments Interface Accounts, Alok Roy

Defensive Publications Series

The present disclosure relates to a secure method and system for re-registration of a Unified Payments Interface (UPI) account, which employs a Cloud-Backed Device Continuity Token (CDCT). The method includes generating the CDCT during initial user registration using a one-way hash function based on a random number, a mobile number, and a timestamp. The CDCT is securely stored in both a backend server of a UPI payment application provider and in a user cloud identity vault. During a re-registration attempt on a new device, the presence and authenticity of the CDCT are validated. If the token matches, re-registration is allowed, …

“Process For Preparation Of Crystalline Tafamidis”, Anonymous

Defensive Publications Series

The present invention relates to a process for the preparation of crystalline tafamidis.

Ai-Guided Controlled Web Exposure Framework For Early Detection Of Web-Based Malware, Chiranthan Niranjan

Defensive Publications Series

The rapid growth of web-based services has significantly increased the attack surface for malware distribution, particularly through non-secure or poorly protected websites. Traditional antivirus and endpoint protection systems are largely reactive, often detecting threats only after widespread infection has occurred. This delay allows emerging malware to compromise a large number of users before effective countermeasures are deployed.

The presented disclosure propose an AI-guided controlled web exposure framework designed to proactively identify emerging web-based malware and evaluate antivirus effectiveness under high-risk conditions. The proposed system operates within fully isolated virtual environments and utilizes multiple controlled administrator-level user accounts to simulate realistic …

System And Method For Emotion-Gated Linguistic Model Training Using Multimodal Emotional Confidence Evaluation, Chiranthan Niranjan

Defensive Publications Series

The presented proposal discloses a system and method for controlling linguistic model training using multimodal emotional confidence gating. The proposed system receives audiovisual media comprising spoken dialogue and associated visual, acoustic, and contextual signals, and automatically segments the media into scene portions based on detected emotional stability characteristics. A multimodal emotion analysis module generates an emotional confidence score for each scene portion by fusing facial expression features, vocal prosody parameters, temporal emotional consistency indicators, and contextual correlation data. A training control module compares the emotional confidence score against a configurable threshold and selectively authorizes or suppresses linguistic parameter updates of …

Delta-Driven Data Rebuilding System With Integrity Verification Via Layered Checksum, Alok Roy

Defensive Publications Series

A system and method for reconstructing and validating versions of a data record. The system continuously monitors and stores data records, data changes and checksums of a plurality of versions of each data record after every operation. Upon receiving a request from a user device to roll back to a target version of the data record the system determines proximity of the target version to first and last versions of the data record and determines traversal direction based on the proximity of the target version. The system generates target version of the data record by traversing through data changes in …