Automated Software Bug Reproduction Via An Iterative Multi-Agent System., Leonid Kuligin

Defensive Publications Series

Resolving software defects can be delayed by incomplete bug reports, which may require manual developer follow-up to obtain information needed for reproduction which increases the time required to fix the bug (mean-time-to-resolve). The disclosed technology describes an automated, iterative multi-agent system that can process these reports. The system can ingest a report (for example, a bug raised by a user through an issue tracker) and use specialized agents (for example, for coding, verification, and reflection) within a reproduction loop to attempt to generate a runnable code example that triggers the described error. A reflection agent can analyze failed attempts to …

Ai-Based Ambient Tutoring Via Generative Correction Of Spoken Language, Alexander Williams

Defensive Publications Series

Some technology-assisted language learning tools may involve structured, explicit study sessions separate from a user's daily activities. This disclosure describes an ambient language tutoring approach. A computing device, such as a smartphone, smart watch, wearable device, or augmented reality glasses, can be placed by the user into a tutoring mode. In this mode, the device uses a microphone to capture spoken utterances from the user in a target language. An artificial intelligence model may analyze the transcribed speech to identify potential grammatical, syntactical, or lexical errors. Based on this analysis, the model can generatively create a corrected, contextually appropriate version …

Intelligent Cybersecurity Training And Response Platform, Rishalin Pillay

Defensive Publications Series

This disclosure describes an intelligent cybersecurity training and response platform that utilizes generative artificial intelligence (genAI). The platform includes a cybersecurity scenario generator that leverages genAI that acts as an expert cybersecurity instructor. The platform also includes an AI-powered investigation coaching engine that provides guidance to analysts. The platform further includes a cross-alert correlation engine to identify complex attack campaigns with high confidence. An evidence collection engine integrates AI-powered analysis directly into the forensic evidence collection process and automatically generates a verifiable chain of custody. The platform creates a seamless lifecycle from training scenario creation to post-incident forensic analysis. The …

Powering Small Devices Using Micro Power Generation, Joel Richard Goergen, Tushara Ramesh, Emily Richter, Yuying Li

Defensive Publications Series

A micro power generation system is proposed herein that can be used to power a small device, such as a wireless access point (AP), to reduce data center power demands for low powered devices, thereby making such devices self-sufficient. The power generation system utilizes air duct flow, batteries, small generators with low starting resistance, and hybrid turbine blades. Tests indicate that such a system can consistently generate 5 watts or more.

Personalized Automatic Query Expansion Suggestions, Na

Defensive Publications Series

While autocomplete suggestions are a feature of search engines and other applications, these are typically limited to a few words and are one-shot. Interactions with a large language model (LLM) can benefit from rich, detailed prompts since the quality of responses provided by the LLM can improve with such prompts. However, such prompts are laborious to write. This disclosure describes techniques that leverage a large language model (LLM) to automatically suggest unstructured, natural-language query completions based on LLM reasoning. With user permission, appropriate user data can be accessed and used for the specific purpose of query completion. The techniques can …

Energy-Cost And Grid-Aware Scheduling Of Computational Tasks On Hardware Accelerators, N/A

Defensive Publications Series

The technology described in this paper relates to scheduling large-scale computational tasks, such as those related to large language models (LLMs), on specialized hardware accelerators. The technology utilizes a decision engine that dynamically evaluates external energy market data, utility grid status, and renewable generation forecasts concurrently with internal task characteristics, including service level objectives (SLOs). By employing a multi-factor cost function, the technology seeks to balance adherence to task deadlines against minimization of operational electricity costs and reduction of strain on the power grid. This allows for high-priority, low-latency requests to be processed immediately, while delay-tolerant batch requests are strategically …

Bias-Corrected Multi-Agent User Feedback Analysis System For Llm Performance Optimization, Saurabh Gupta, Andrea Morandi

Defensive Publications Series

Proposed herein is a novel bias-corrected multi-agent user feedback analysis system that automatically corrects for selection bias in sparse user feedback to provide accurate, unbiased quality assessments of Large Language Model (LLM) systems.

Optimizer System And Method, Sayantika Sarkar, John H. Li, Shari Krikorian, Raymond B. Choi

Defensive Publications Series

The present disclosure provides a system and method for managing the payment of invoices using an optimizer system. The method comprises receiving, by an optimizer system, an accounts receivable file comprising a plurality of invoices from an accounts payable system, the accounts payable system configured to process invoices provided by a supplier to a buyer for payment; analyzing, by the optimizer system, the plurality of invoices in the accounts receivable file; determining, by the optimizer system, one or more invoices in the plurality of invoices to accelerate for payment; transmitting, by the optimizer system, a first message to the accounts …

Novel Crystalline Form Of 6-[(6,7-Dimethoxyquinazolin-4-Yl)Oxy]-N,2- Dimethyl-1-Benzofuran-3-Carboxamide And Its Process For The Preparation, Msn Laboratories Private Limited, R&D Center; Srinivasan Thirumalai Rajan; Sagyam Rajeshwar Reddy; Gade Srinivas Reddy; Kammari Balraju; Bingi Hariprasad Reddy; Padigapaty Shiva Reddy And Matapathi Mahesh.

Defensive Publications Series

This application relates to a novel crystalline form of 6-[(6,7-dimethoxy quinazolin-4-yl)oxy]-N,2-dimethyl-1-benzofuran-3-carboxamide, also known as Fruquintinib. The disclosure also covers the process for preparing this new crystalline form.

Formula-1.

The present disclosure also relates to a process for the preparation of compound of formula-9 which is a key intermediate used in the preparation of Fruquintinib.

Formula-9

The present disclosure also relates to a process for the preparation of Fruquintinib.

Event-Triggered Audio Mode Management For Hearable Devices, Shailesh Maheshwari, Vedant Maheshwari, Rayansh Maheshwari

Defensive Publications Series

This document describes a method for managing audio modes in a hearable device. A user operating a device in an ambient sound passthrough mode may have difficulty hearing audio events, such as notifications or digital assistant responses, over background noise. A system can detect certain triggers, such as the receipt of a notification or the start of a user's interaction with a digital assistant. Upon detecting a trigger, the hearable device can temporarily switch from an ambient sound mode to a noise cancellation mode. After the audio event concludes, for example, after the notification sound or the assistant's response is …

Structured Layered Synthesis For Generating Coherent And Scale-Consistent Text, Saurabh Gupta, Andrea Morandi

Defensive Publications Series

Although Large Language Models (LLMs) excel at generating fluent text at a local level, LLMs struggle to consistently maintain the structure and coherence of generated text over long documents and across different levels of detail. Techniques proposed herein introduce a system and method for enhancing text generation through a structured layered synthesis approach.

Stateful News Ranking To Prioritize Novel Content For Repeat Queries, Deep Shah, Aditya Bhandari

Defensive Publications Series

A method for stateful ranking of search results may be used to prioritize novel content when a user repeats a query, for example, for a developing news story. A system can establish a "context timestamp" based on a user's most recent prior query on a topic. The system may then identify "context results," which can be items published before this timestamp, and "latest results," which can be items published after it. To differentiate new information, the system may computationally compare the latest results to the context results, for instance, by generating vector embeddings and identifying latest results with a similarity …

Generating Context-Aware Code-Switched Text Using Probabilistic Modeling And Generative Refinement, Mithun Kumar S R

Defensive Publications Series

Artificial intelligence models can experience difficulty understanding and generating code-switched text, and some existing synthetic data generation methods can produce grammatically incorrect or unnatural-sounding results. This disclosure describes a multi-stage system for generating code-switched text from monolingual corpora. The method can employ a probabilistic constraint model to analyze a source sentence and identify syntactically and semantically appropriate points for a language switch. Based on these identified points, a candidate generation engine can create one or more raw code-switched sentences. A large generative model may then refine these raw candidates to address potential grammatical errors at the language junction and improve …

System And Method For Topologically Protected Quantum Computing Based On Majorana Fermions In Superconducting Nanostructures With Integrated Cryogenic Control, Lorenzo Lbn Bernardini, Origines Technologies

Defensive Publications Series

The present invention is located in the field of topologically protected quantum computing, with particular reference to the implementation of qubits based on Majorana fermions in nanoscopic superconducting structures. This technological field represents an innovative convergence between condensed matter

in physics, quantum computing and superconducting materials engineering. The system object of the present invention operates in a technological domain where quantum hardware based on Majorana fermions is integrated with sophisticated topological protection mechanisms of quantum states. This integration is achieved through the implementation of qubits that exploit topologically protected states within nanoscopic superconducting structures, allowing a coherent manipulation of non-abelian …

On-Device Large Language Model Inference With Unlimited Context Size, Xusong Wang, Miao Wang

Defensive Publications Series

On-device large language model (LLM) pipelines that employ hardware accelerators frequently encounter limitations regarding the maximum context size that can be processed on-device. This constraint arises because neural network accelerators are designed to favor the execution of statically-shaped graphs, whereas varying context sizes necessitate dynamic dimensions within the attention layer of a neural network. Conventional solutions present challenges such as computational inefficiency and lack of scalability. This disclosure addresses the issue of context size limitation when performing inference using an on-device LLM by partitioning the attention graph into multiple reusable, statically-shaped graphs that are mathematically equivalent to the original computation. …

A Modular Pipeline For Flatbed Content Detection And Localization With Accurate Deskew, Hp Inc

Defensive Publications Series

We present an end-to-end pipeline for localizing and deskewing one or multiple foreground objects (documents, photos, cards, receipts) placed on a flatbed scanner with a white to near-white background. The method integrates iterative morphological contrast shaping (white / black top-hats), gamma-driven diffusion, statistical region merging (SRM), distance-transform based consolidation, and Voronoi-like label propagation. A unified skew estimation interface produces deskewed, refined crops per object. This document mirrors the structural flow of a companion skew orientation disclosure while remaining self-contained and focused on the flatbed multi-object extension.

Outdoor Optical Distribution Frame, N/A

Defensive Publications Series

The technology described in this publication relates to a high-density, weather-proof optical distribution frame (“ODF”) that enables network setup outdoors in parallel with data center construction, designed to accelerate machine learning (ML) data center capacity delivery. This frame eliminates the need to wait for lengthy building completion before initiating network deployment. By enabling earlier network readiness, this frame has the potential to shorten overall data center delivery timelines by 3 to 6 months.

Leveraging Large Language Models For The Analysis Of Network Security, Claudio Criscione

Defensive Publications Series

Developing security scanners to identify vulnerable software within the network of an organization can be expensive and error prone: for each new software or version thereof, a security engineer deploys a live instance and determines fingerprintable properties that can be automatically identified. This disclosure describes techniques that leverage large language models (LLMs) to perform attack surface management functions such as the discovery of software that is exposed to attackers and the vulnerabilities of such software. A reconnaissance scan of the network perimeter determines HTTP endpoints within the network. The HTTP endpoints are fed to the LLM alongside a prompt instructing …

On-Demand Generation Of Searchable Ephemeral Web Addresses Via Agentic Browser Actuation, Tal Snir

Defensive Publications Series

Traditional web crawlers may face challenges indexing certain types of dynamic, ephemeral web content, such as customized product comparisons or filtered lists, as this content can be generated on-the-fly and may not have a persistent, static uniform resource locator (URL). This situation can lead to search results that do not include certain types of user-customized information. A disclosed system can be configured to capture user intent from a query and deploy an agentic technology, for instance, a browser actuation agent. The agent can programmatically simulate user navigation on a target website to dynamically generate an ad-hoc URL corresponding to content …

Decoupled Image Retrieval To Augment Large Language Model Output, Pulkit Jain, Josh Ain, Sowmya Vijayaraghavan, Daniel Hirsch, Jason Fund

Defensive Publications Series

Augmenting textual responses from large language models (LLMs) with relevant images can present challenges at scale, for example, due to computational resources and latency that may be associated with some integrated multimodal models or co-embedding techniques. The disclosed method can address this by classifying a user query, for instance, to determine if it is "image-seeking." If a query is identified as image-seeking, the system can forward the query and associated metadata to multiple systems that can operate in parallel. For example, the query can be sent to an LLM for generating a text response and also to a separate image …

Coordinated Agent System For Automated Market Viability Assessment Of Product Designs, Ankit Aggarwal, Piyush Dhanuka

Defensive Publications Series

Some product design processes can involve time-consuming and costly manual research for market feasibility analysis, potentially increasing financial risk. A described agentic system is configured to assist in automating this process. The system may include a central orchestrator agent, powered by a large language model, that can receive a product design, such as an image, along with target parameters. The orchestrator can then invoke specialized services and parallel retrieval-augmented generation tools. These tools are configured to query distinct knowledge bases to gather data on topics such as market trends, competitive pricing, cost of goods, and supply chain logistics. A financial …

Improving Large Language Model Responses Using Implicit Feedback From Conversational Refinement Metrics, Tushar Chugh, Aditya Mone, Mehrbod Sharifi, Karthik Kumara

Defensive Publications Series

Techniques for training large language models (LLMs) rely either on sparse, explicit user feedback or expensive human evaluation. These techniques fail to capture the nuanced conversational friction that users experience during multi-turn dialog. This disclosure describes techniques that, with user permission, leverage follow-up prompts entered by a user as a dense, scalable source of implicit feedback for post-training LLMs. A refinement framework classifies conversational turns following an LLM response based on the relationship between the turns. The framework categorizes user prompts into two classes - refinement (a negative signal indicating user friction and an attempt to fix a failure in …

(((Qpie))) Prior Art 6 Of 9 Aetheris Economics – Why The Physics Of Coherence Becomes The Business Model Of Trust, Teddy Burroughs

Defensive Publications Series

(((QPIE))) Prior Art 6 of 9 Aetheris Economics – Why the physics of coherence becomes the business model of trust

“What does Aetheris actually mean for revenue, social impact, and SLA innovation?”

AETHERIS AND THE ECONOMICS OF RESONANCE

Why the physics of coherence becomes the business model of trust

1. The Shift From Efficiency to Coherence

For the past century, growth came from efficiency: doing more work with fewer inputs. Buqt efficiency has a hidden limit—it assumes the environment is static and controllable. In a post-quantum, post-pandemic, climate-entangled world, stability comes not from control but from coherence—the capacity of …

Multi-Channel Ambient Sample System (Mass): A Flush-Mounted Multichannel Microphone Array For Hemispherical And Spherical Spatial Sound Capture, Stefan Johansson

Defensive Publications Series

This document discloses a method and apparatus for recording spatial sound using a multichannel microphone array based on the acoustic principles of the Crown® SASS (Stereo Ambient Sampling System), but extended to cover the entire 360° horizontal plane and optionally the upper hemisphere. The disclosed system, named MASS (Multi-channel Ambient Sample System), employs flush-mounted omnidirectional microphone capsules arranged on multiple planar surfaces of a cabinet, with acoustic baffling to control interchannel separation and diffraction. This disclosure is intentionally placed in the public domain to establish prior art, thereby preventing others from obtaining patent protection for this class of …

Server-Side Stitching Using A Vendor Identifier For Application Conversion Attribution, Ishan Bansal, Satvik Chauhan

Defensive Publications Series

In some mobile app conversion tracking scenarios, attribution parameters on post-install engagement (PIE) events may be inconsistently transmitted, leading to potential gaps in measurement. A method is described to address this using a server-side stitching mechanism that can utilize a vendor identifier (e.g., an identifier for vendors or an application set identifier). During an initial app landing event where attribution parameters are present, a subset of data from these parameters may be stored on a server and keyed using the vendor identifier. For subsequent PIE events from the same app that lack these parameters, the system can use the vendor …

Automated Contextual Selection And Placement Of Emojis In Advertising Headlines, Alan Ruthazer, Parag Jain, Haifeng Gong, Wei Yin, Thomas Palmer

Defensive Publications Series

Advertising platforms may restrict emojis in ad headlines due to challenges related to contextual relevance, scalability, and ad safety. This disclosure details a method for the contextual selection and placement of emojis within ad headlines, for example, by using a generative artificial intelligence model. A system, which can be executed on a computing device (e.g., a server, a smartphone, a wearable device), may ingest ad headline text and perform a deep learning analysis to understand its core meaning and context. Based on this analysis, the system can select and rank relevant emojis from a repository. The technology may also identify …

Ai-Driven Techniques To Dynamically And On-Demand Provide Context-Aware Task Optimization In Complex Systems, Arica Schivonne Jackson, Priya Santosh Singh, Angeliki Papathanasiou, Natallia Shliazhka, Sebastian Jeuk

Defensive Publications Series

In today's fast-paced and digitally driven environment, professionals are bombarded with many different types of notifications on a daily basis in which such notifications can be scattered across different platforms, such as email platforms, instant messaging platforms, and the like. Notifications spread across different platforms can lead to a loss of productivity due to wasted time and duplicated effort. To address such issues, an intelligent, centralized system is proposed herein that aggregates artificial intelligence (AI) -driven insights and automates task management across multiple platforms, providing a cohesive and streamlined user experience. Such a system can reduce cognitive load, enhance productivity, …

Sharer Intention Analysis For Content Type Detection, Jieyu Tan, Karlheinz Wurm, Soya Li, Wallice Wang

Defensive Publications Series

In online conference scenarios, shared content detection is widely used to determine whether content being shared is text content or video content. Determining the type of shared content can impact whether the video mode for the online conference favors higher resolution (for text content) or smoother motion (for video content). A system is proposed herein that leverages predictive hints from user actions (e.g., opening applications or files) and contextual audio cues to anticipate upcoming shared content types (e.g., video or text). By combining these hints with real-time content analysis, the system preemptively switches to the appropriate video mode for an …

Automated Software Service Onboarding And Configuration Through A Conversational Large Language Model Agent, Peng Chen

Defensive Publications Series

Software service onboarding can involve developers manually interpreting integration guides to edit various configuration files, a process that can be time-consuming and prone to error. A system is described that may use a conversational agent powered by a Large Language Model (LLM) to assist in automating this process. The agent can interact with a developer, for example through a user interface on a computing device, to gather parameters. These parameters may then be populated into a templatized prompt for the LLM to generate corresponding configuration files, code snippets, and commands. The system may integrate with existing development frameworks and could …

Proactive And Semi-Autonomous Agentic Quality Management Framework, Ramakrishnan Neelakandan, Teginder Singh, Bakul Patel

Defensive Publications Series

Traditionally, quality management relies on siloed systems of record such as quality management system (QMS), application lifecycle management (ALM), and manufacturing execution system (MES) platforms. These systems are often static, passive repositories that require significant manual effort to connect disparate data and derive actionable insights. Fragmentation and lack of proactive intelligence can lead to delays in identifying quality issues, ensuring compliance, and accelerating innovation. This disclosure describes a quality management framework to provide collaboration between human experts and specialized artificial intelligence (AI) agents for proactive and semi-autonomous quality management. The framework provides a distributed, intelligent ecosystem where a central AI …