Mukesh Reghu

About Mukesh Reghu

Problem Solver | LLM Enthusiast | Deep Learning for NLP | Deep Learning | Explainable AI | Machine Learning | Keen Learner | AWS Certified Solutions Architect

Experience

1. Sr Ai Scientist

   Uniphore

   Aug 2024 — Present · Chennai, IN

   Built and productionized a new multilingual intent detection system for a real-time agent assist product, replacing a legacy system in production. Benchmarked the intent detection system against the incumbent across accuracy, latency, concurrency, and cost using internal customer POC data and public datasets; presented results to product managers and stakeholders. Led production integration end-to-end for the intent detection system, supporting QA, release, and enabling sales and delivery teams with customer demos. Extended the intent detection system system to multiple languages and applied LLM-based techniques to accelerate training and test data creation. Designed and evaluated LLM-based approaches for predicting the disposition path/leaf for a given call from a tree of hierarchical dispositions, such as recursive prompting and candidate retrieval followed by structured reasoning, achieving 47% exact-match accuracy on a customer dataset. Extended a production RAG system to support Arabic by benchmarking embeddings, re-rankers, prompts and LLMs using language experts and LLM-as-a-judge. Improved the semantic retrieval in the system by over 10%. Built an LLM-based canonicalization system to normalize semantically equivalent texts into a single canonical form, enabling reliable grouping, reporting, and insights; handled cross-pod race conditions using Redis-based distributed locks.

Education

• Indian Institute of Information Technology Design & Manufacturing Kancheepuram

  Dual Degree (B.Tech. + M.Tech.), Computer Engineering

  2015 — 2020