-
Enhanced automatic speech recognition to avoid misrecognition of voice utterances (US Patent 12,525,220, 2026)
Describes a post‑ASR model that flags likely misrecognitions, selects a corrected interpretation, and routes it to the response-generation system.
-
Where Should Intelligence Live? Agent-Centric vs. Environment-Centric Design in Agentic Systems (Authorea Preprints, 2026)
Practical framework for choosing between general-purpose agents and specialized systems; argues specialization wins in structured/high-reliability settings and hybrid designs mitigate common failure modes.
-
From “Everything is a File” to “Files Are All You Need”: How Unix Philosophy Informs the Design of Agentic AI Systems (arXiv:2601.11672, 2026)
Connects Unix’s “everything is a file” abstraction to agentic AI, proposing file/code-centric interfaces that make tools and resources more composable, auditable, and robust.
-
Dynamic LLM routing and selection based on user preferences: Balancing performance, cost, and ethics
Introduces a routing engine that selects the best LLM per request using user-defined tradeoffs (accuracy, latency, cost, and ethics) instead of defaulting to the largest model.
-
The Unification Imperative: End-to-End Models Are Replacing Modular Pipelines in AI Systems
-
The AI Roles Continuum: Blurring the Boundary Between Research and Engineering
Synthesizes evidence from AI org hiring signals to argue “research” and “engineering” roles increasingly overlap, proposing an AI roles continuum across common archetypes.
-
PROFASR-BENCH: A Benchmark for Context-Conditioned ASR in High-Stakes Professional Speech
Presents a professional-speech benchmark across finance/medicine/legal/tech, emphasizing entity-critical errors and controlled evaluation of context-conditioned ASR.
-
Mind the Goal: Data-Efficient Goal-Oriented Evaluation of Conversational Agents and Chatbots using Teacher Models
Goal-oriented evaluation for multi-turn chatbots: measures whether the user’s end goal is achieved and provides a taxonomy of root causes when conversations fail.
-
Large language model (LLM)-based correction based on a multi-tool prompt (US Patent 12,444,412, 2025)
Covers an LLM-guided workflow where a prompt encodes a multi-step tool plan; the model generates tool inputs and applies tool outputs to iteratively update user-provided data.
-
Automated ASR Transcriptions LLM Agent Learning to Imitate Humans (Zenodo, 2023)
-
Selfplay: A Python Framework for Role-Based Dialogue Simulation (2024)
-
Utterance generation and evaluation (US Patent 11,600,260, 2023)
Techniques for generating candidate utterances from intent/target data and automatically evaluating them for quality and suitability in downstream voice/NLU workflows.
-
Voice-based Computer Interfaces for Natural Language Inputs, Patent, Filed 2022
-
Customer-assisted Recovery of ASR errors using customer enunciations in Alexa, Patent, Filed 2024
-
Interactive ASR error recovery using Reinforcement Learning, Patent, 2024 - Status Unknown
-
WEB2VOICE - Bridging the gap between Voice Shopping and Web Shopping to Improve ASR, Amazon Machine Learning Conference (AMLC) 2022
-
Retrieve-Rank - ASR Error Correction for Long-tail Domains, Amazon Machine Learning Conference (AMLC) 2022
-
Identifying Product Affinity in Webpages for Contextual Advertising, Amazon Machine Learning Conference (AMLC) 2018