Layout-Aware Text Extraction Using Heuristic Segmentation And Llm-Based Refinement, Leonid Kuligin

Defensive Publications Series

Automated text extraction from paginated documents, such as portable document format (PDF) files, can be problematic when layout-agnostic parsers capture non-essential content like headers and footers, which can introduce noise and degrade data quality for downstream applications. A described technique may address this through a multi-stage workflow that combines heuristic analysis with large language models (LLMs). For example, a process can begin with a heuristic segmentation stage to identify potential header, footer, and main content regions based on layout cues, such as text position and font size. This initial segmentation may then be provided as a structured hint to a …

System For Jurisdictional Mapping And Dynamic Routing Of Cloud Api Requests, Johannes Start, John Lunney

Defensive Publications Series

Cloud agents interacting with third-party services via generic application programming interface (API) endpoints should adhere to data residency requirements. A system may intercept outbound API requests to help manage data sovereignty policies. The system can use a hierarchical region-to-jurisdiction map and a service capability matrix to determine a request's source jurisdiction and identify compliant endpoints for the destination service. If a compliant endpoint is available, the system can dynamically rewrite the request to target that endpoint. If no compliant endpoint is found, the request can be blocked. This technique may facilitate the enforcement of jurisdictional data processing rules by decoupling …

Geographically Constrained Ai Agent Execution Via Signed Policy Tokens, Johannes Start, John Lunney

Defensive Publications Series

Systems for managing data sovereignty can present challenges in governing autonomous artificial intelligence agents, particularly regarding downstream network calls that agents may generate dynamically. This situation can introduce data residency compliance challenges. A described system can attach a portable, cryptographically signed policy token to a data processing request. This token, which may be derived from a residency profile, can specify a geographical or jurisdictional boundary for data processing. Enforcement middleware within an execution environment, such as a cloud computing platform or a containerized environment, can intercept an agent’s outbound network calls and check the destination against the policy contained within …

Orchestrating Agent Failover Using A Relationship Tree With Defined Dependency Types, Gaurav Sangamnerkar, Himesh Mukherjee

Defensive Publications Series

In distributed computing environments with multiple interacting agents, interdependencies can contribute to localized agent failures escalating into system-wide outages, as some protocols may not adequately define dependency-based failover strategies. A control plane framework can manage such interdependencies by employing an agent relationship tree (ART). The ART can be a data structure that maps the relationships between agents and annotates each with a dependency type, for example, solid for dependencies designated as strongrelationship or fluid for those with lower importance. An agent orchestrator can monitor agent health and, upon detecting a failure, can consult the ART to initiate a context-aware recovery …

Downhole Tools Thermal Shield

Defensive Publications Series

Measurement and Logging While Drilling (M/LWD) tools encounter significant challenges when drilling in high-temperature (HT) oil and gas or geothermal wells. These extreme environments can cause downhole tools to fail and sensors to drift, leading to inaccurate readings and compromising the safety and efficiency of drilling operations. Sensitive electronic components, such as sensors, wiring, and circuit boards, are designed to operate within specific temperature ranges, and prolonged exposure to elevated temperatures can result in malfunction or reduced lifespan.

Secure Decommissioning Of Cloud Resources, Pardha Saradhi Mavilla

Defensive Publications Series

Soft deletion is a feature in cloud computing that provides a safety net against accidental deletions using a predefined grace period within which the cloud resource is recoverable. However, soft deletion creates new operational bottlenecks for users who intended to permanently delete their resources, e.g., continued resource occupation by soft-deleted clouds, inhibited redeployment of resources, etc. This disclosure describes techniques for secure soft deletion of cloud resources in a manner that fulfills operational, security, and user experience requirements such as accelerated decommissioning, visibility into the state of decommissioning, and deterministic rollback of decommissioning. A robust, end-to-end, orchestrated procedure balances resource …

Dynamic Antenna Configuration For Ntn Links Via Predictive Look-Up Table Search, Ali Molaei, Kasra Ghaemi, Gregory Black

Defensive Publications Series

Establishing communication links between mobile terminals, such as smartphones or vehicle communication systems, and non-terrestrial networks (NTNs), for instance, low earth orbit satellite constellations, can be impacted by frequent variations in signal characteristics as satellites move. If a static antenna configuration on a mobile device does not adapt to this dynamic signal environment, link degradation may result. A system for dynamic antenna configuration can use a pre-calculated look-up table (LUT) that maps various satellite positions and signal properties to corresponding antenna tune states. During operation, a predictive algorithm on the terminal can use real-time signal strength measurements and the LUT …

Current Sharing Method For Digital Voltage-Mode Power Converters, Na

Defensive Publications Series

The technology described in this paper relates to a current sharing method for digital voltage-mode power converters connected in parallel. Specifically, the disclosed technology introduces a current sharing loop that calculates the discrepancy between the local output current and the average current of the power stages and adjusts the voltage loop integrator value based on this error. By modifying the integrator state directly, the system enables immediate duty cycle correction without increasing the control system’s order. This approach achieves rapid and stable current sharing within the switching period while maintaining high voltage regulation accuracy without the need for traditional droop …

Technical Evaluation Of The Interactive Digital Vinyl System (Idvs), Christoph Fussgänger

Defensive Publications Series

This technical evaluation presents the Interactive Digital Vinyl System (IDVS) and the IDVS+, a sensor-based Digital Vinyl System (DVS) control vinyl that adds a rotating touchscreen interface for direct, on-deck control of DJ and music software.

The IDVS+ is a 12-inch, battery-powered device that operates without a stylus, continuously measures its own rotation via internal motion sensors, and transmits motion data wirelessly to a connected computer, DJ player, or smartphone, where the audio is generated and played back. In addition to standard DVS behavior (speed control, forward and reverse movement, and scratching), the IDVS+ introduces a high-resolution touchscreen surface capable …

Framework For Situated Agents Using Tool-Based Perception And Interaction In A Continuous Cognitive Loop, Valentyn Shybanov

Defensive Publications Series

Agentic systems that operate in a transactional or stateless mode can be limited in their ability to perform sustained, multi-step tasks within dynamic digital environments. Such systems may exhibit brittleness, where tool execution failures may lead to a process halting rather than initiating an adaptive recovery. The disclosed technology describes a framework for autonomous agents centered on a continuous perception-cognition-interaction loop. Within this framework, an agent can utilize programmatic tools as sensory inputs to perceive the state of a digital environment and as actuators to effect change, while a cognition layer can process these perceptions to formulate actions. This cyclical …

System For Ai-Driven Ui Control Via A Hardware Actuator Emulating A Human Interface Device, Shashwat Razdan

Defensive Publications Series

The ability of intelligent agents to perform complex, cross-application tasks may be constrained by a dependency on application-specific programming interfaces (APIs), which can limit interaction with user interfaces and adaptation to dynamic visual changes. A system can employ a closed-loop perception-action cycle where a multimodal intelligent agent analyzes visual data from a screen of a host device, such as a smartphone or computer, to predict a next action. A control application can translate this action into a sequence of low-level human interface device events. A peripheral hardware actuator device may receive these events and inject them into the host device's …

Conditional Execution Of Compiler Passes Based On Per-Script Feature Detection, Rishipal Singh Bhatia, Laura Harker

Defensive Publications Series

Source-to-source compilers may perform inefficiently by executing transpilation passes on scripts that do not contain the specific language features a pass is designed to transform, potentially leading to redundant processing. A compiler can analyze a script to generate a per-script feature map, for example, by identifying language features in its abstract syntax tree (AST). Before executing a transpilation pass, the compiler can check this map and may bypass the pass for that script if the specific feature targeted by the pass is not present. This feature map can also be dynamically updated throughout the compilation process as other passes transform …

Multiple Exposure Phase Detection Blending For Autofocus Cameras, Sunghyun Hwang, Ryan Chan, Dajung Lee

Defensive Publications Series

This disclosure describes autofocus (AF) techniques that leverage multiple exposures of phase detection (PD) data to ensure that all relevant regions of interest in an image benefit from the highest possible PD signal quality while reducing the computational load and latency incurred by the AF module. A single, unified PD raw frame is generated by merging data from multiple exposure contexts. Raw PD data is blended within the front-end of the image signal processor (ISP) or similar low-level layer, obviating extensive preprocessing to determine the optimal PD source. Enhancement in autofocus quality is achieved especially for dark objects in high …

Llm Specific Video Ad Format, Caroline Pieracci

Defensive Publications Series

Disclosed herein is a system and format for integrating video advertising natively into Large Language Model (LLM) conversational interfaces. The format consists of a content package including a short, educational video bundled with a rich metadata payload. The payload comprises a verified transcript, source citations, and an authority score indicating reliability. A verification engine analyzes submitted content for factual accuracy, non-promotional tone, and production quality to assign the authority score. Keywords: LLM advertising, conversational AI, native video ads, authority score, reliability, sponsored answers, verified content.

Technical Addendum: Autonomous Thermal Balancing & Closed-Loop Architecture For Allfromair™ (Gen 3), Vesa Olavi Lius

Defensive Publications Series

This document serves as a formal Technical Addendum to the ALLFROMAIR™ Unified Thermodynamic Platform. It discloses the proprietary "Terra" configuration, a system architecture designed for autonomous Atmospheric Water Generation (AWG) and Tri-Generation in inland, arid, or industrial environments lacking natural thermal sinks (e.g., sea or lakes).

The addendum details the engineering of an Internal Hydrological Loop coupled with a thermal accumulator buffer. Key disclosures include:

* Closed-Loop Thermodynamics: A method for recirculating process water to achieve Zero Liquid Discharge (ZLD) while balancing thermal loads internally.

* Active Thermal Balancing: The utilization of the Third Stage (SVRU 2) as a dynamic …

Agentic Workflow For Translating Natural Language Intent Into Validated, Structured Business Rules, Marim Ganaba, Jeannette Gatlin, Jarred Smeltzer, Dearbhla Moore, Alex Pineda

Defensive Publications Series

Conventional enterprise systems for creating business rules, for example sales compensation plans, can involve manual translation of strategic intent into structured configurations. This process may place a cognitive burden on users. The disclosed technology describes a system employing an orchestrated, multi-agent workflow to assist in this translation. A planning orchestrator agent can receive a user's intent in natural language, deconstruct the goal, and delegate sub-tasks to specialized agents. These agents may perform functions such as knowledge retrieval, parameter synthesis using retrieval-augmented generation, and compliance validation. The workflow can generate a validated, structured data object, such as a javascript object notation …

Multi-Agent System For Interactive Modeling And Lifecycle Management Of Business Rules, Marim Ganaba, Jeannette Gatlin, Dearbhla Moore, Jarred Smeltzer, Allen Kempson, Kenneth Ishikawa, Alex Pineda

Defensive Publications Series

Certain methods for managing complex business rules, such as those in enterprise performance management, may rely on fragmented tools that can lack the capability to model cross-domain impacts of proposed changes, which can lead to siloed decision-making and manual governance processes. Systems and methods are described that can utilize a multi-agent architecture to provide an interactive simulation environment, or sandbox, for business rule lifecycle management. Specialized software agents, for example data quality, planning, governance, and triage agents, may collaborate to support tasks from initial data validation to automated testing and deployment. A simulation engine can provide feedback on the cascading …

Hierarchical Pre-Decoding Reliability Verification Of Generative Model Latent States, Neel Mohbe

Defensive Publications Series

Reliability verification for generative artificial intelligence models may be constrained by a reliance on post-hoc, token-level analysis, which can be computationally inefficient. Described are systems and methods for a pre-decoding verification architecture that can operate on the internal latent states of a generative model before output tokens are sampled. The architecture can use a hierarchical cascade of checks, for example, beginning with a fast linear classifier, escalating ambiguous inputs to a non-linear density-based evaluator, and routing complex cases to specialized expert models. A feedback mechanism can propagate classifications from the expert models backward to update and refine the faster, earlier …

In-Line Gasless Fluxing, John Robert Buster Mccallum, Eric Michael Ball, Jeffrey M. Mcdermott, Jocelyn Rose Cracroft

Defensive Publications Series

In view of the limitations of existing inline degassers utilizing salt, described herein are inline degasser systems and methods which introduce salt utilizing various mechanical methods and/or systems in place of the inert gas.

An Improved Process For The Preparation, Isolation And Purification Of Dual Agonist Polypeptide, Anonymous

Defensive Publications Series

An improved process for the preparation, isolation and purification of dual agonist polypeptide compound.

Online Platform Or Service Access Tiers Based On Data Permissions, Na

Defensive Publications Series

Many software applications and online services provide free or limited usage access to users. Such products and services rely on commercial revenue from advertising or other mechanisms that rely on usage of user data. To use such services, users provide their consent for limited use of data such as activity data on the platform for commercial purposes. This disclosure describes techniques that offer to compensate users for their data that is specific to the usage of an online service or platform, and is used in accordance with platform policies and applicable regulations. The techniques enable user-selection of an appropriate level …

Browser Engine System For Dynamic, Query-Guided Synthesis And Injection Of Contextual Session Essence Into Large Language Model Prompts, Abhay Prakash, Sophie Chang

Defensive Publications Series

A system and method for dynamic, query-guided synthesis and injection of a contextual session essence into large language model prompts provide a browser-native solution to computational latency and context window saturation. The system comprises a context gathering module that performs real-time state measurement of Document Object Model mutations and session logging, contingent upon a user-authorized processing flag. A context analysis and inference engine transforms the resulting raw data cluster into a high-density natural language summary, referred to as the contextual session essence, by executing on-device machine learning inference for intent classification and entity extraction. The query augmentation module intercepts user …

An Improved Process For Preparation Of Vonoprazan Fumarate., Anonymous

Defensive Publications Series

The present publication relates to a process for preparation of Vonoprazan fumarate of Formula I.

Process For The Preparation Of Uridine Triacetate, Anonymous

Defensive Publications Series

The present invention provides a process for the preparation of Uridine Triacetate

Long-Haul Vehicle With Railcar-Based Range Extender, Mordechai Teicher

Defensive Publications Series

A vehicle-on-railcar system is disclosed for long-haul railway transport, in which a railcar is equipped with a range-extending onboard battery. While an electric vehicle is mounted on the railcar, the railcar battery selectively supplies direct current power to the vehicle inverter, which converts the power to regulated alternating current for driving a railcar traction motor. By supplementing or replacing energy drawn from the vehicle battery, the railcar-based battery enables extended or non-stop long-haul operation without requiring intermediate vehicle charging stops or oversized vehicle batteries. Charging and energy replenishment are shifted to railcar-only operations within railway yards or sidings, improving vehicle …

Technical Disclosure: Rme-1 Resonant Manifold Encoding, Samuel J. Church

Defensive Publications Series

TECHNICAL DISCLOSURE: RME-1 RESONANT MANIFOLD ENCODING

Document ID: RME-2026-001-DISC Publication Date: January 18, 2026 Field: Artificial Intelligence / Geometric Logic Constraints / Hallucination Mitigation

1. Abstract

This disclosure describes a system and method for constraining Large Language Model (LLM) generation through a 3D Geometric Manifold. Unlike traditional probabilistic token selection, the RME-1 protocol enforces structural integrity by measuring the "Reality Gap" between abstract symbols and grounded axioms, utilizing a "Literal-Floor Anchor" to prevent semantic drift.

2. Technical Field

The present disclosure relates generally to the fields of natural language processing (NLP) and machine learning, specifically focusing on deterministic governors that …

Sanitizing Dynamic Strings From System Logs To Enable Comparison Between Nominal And Failing Runs, David Murphy, Noah Kersey, Angelica Rosenzweig Castillo

Defensive Publications Series

The technology involves debugging and testing of changes to operating systems. Aspects of the technology include replacing every dynamic string in a log with a placeholder value, and deterministically assigning the same tags or placeholders across consecutive runs. Lines can be diff'ed to determine their novelty in the triage process. The display of log pairs may be streamlined to hide all lines except the unique lines, which is beneficial in triaging build failures.

System For Managing Ai-Generated Interactive Modules With A Sandboxed Container, Nathan Grabaskas, Fredrik Ihre, Michael Bausor, Megan Pusey, Levi Zombori, Liam Roche

Defensive Publications Series

Digital platforms designed for static content may face challenges integrating dynamic, AI-generated interactive micro-applications in a secure manner. This disclosure describes a system for managing these modules using an interactive module container (IMC). The IMC can function as a sandboxed architectural layer that encapsulates a module's code, data, and state. The IMC may mediate interactions with backend services and a host environment through an abstracted application programming interface. The system may also feature a partitioned data architecture to manage private user data and creator-controlled data with granular permissions. This approach can provide a framework to facilitate the creation, distribution, and …

Integration Of Direct Liquid Cooling With Heterogeneous Packages, N/A

Defensive Publications Series

Integration of Direct Liquid Cooling with Heterogeneous Packages

Abstract

The disclosure describes a hybrid cooling architecture designed for heterogeneous die packages. In the disclosed systems, high-power dies amenable to direct cooling can be liquid-cooled, while other, relatively low-power dies, can be cooled indirectly. As described herein, a novel cold plate architecture can be used to manage flow routing between direct and indirect cooling zones. In addition, in some embodiments, a sealant can be used around the periphery of the dies that receive DLC.

Navigational Path Reversion Analysis From A Prior Divergence Point, Ashutosh Srivastava

Defensive Publications Series

Vehicle navigation systems may calculate prospective routes from a user's current location but may not evaluate the potential benefit of turning around to a prior divergence point. A system can address this by storing data for an unchosen alternative path and its associated divergence point from a selected route. The system can monitor real-time conditions, such as traffic congestion, for both the active route and the stored alternative path. If a trigger condition is met, for example, a significant delay on the active route, a reversion calculation can be performed. This computation can determine an estimated time to backtrack to …