Sudheer koki

About

Built and productionized an LLM-powered internal knowledge search system in a regulated environment, using embeddings/vector DB retrieval with strict grounding and confidence gating to reduce hallucinations. Reported ~45% accuracy improvement over keyword search and implemented end-to-end orchestration, monitoring, CI/CD, and incremental re-indexing to manage latency and data freshness while driving adoption with business stakeholders.

Experience

Education

Key Strengths

• Built and deployed a production LLM-powered internal document search system (RAG)
• Improved answer accuracy ~45% and significantly reduced manual research time
• Strong hallucination control for regulated environments (grounded responses, confidence gating, similarity thresholds)
• Reliability-focused evaluation approach (curated tests, edge cases, measurable metrics: success rate, accuracy, latency, retrieval relevance)
• Production operations mindset (monitoring/logging of retrieval scores, prompts, responses; continuous tuning)
• End-to-end delivery with APIs, CI/CD, and orchestration for repeatable deployments
• Effective collaboration with non-technical stakeholders using prototypes and business-facing metrics

Reference Highlights

• strong technical lead and collaborator
• successfully ships end-to-end AI/data products to production
• strong ownership of production readiness
• delivers measurable business impact with real users
• exceptional cross-team collaboration
• dependable and proactive
• translates complex LLM concepts into business-focused language
• patient and structured with non-technical stakeholders
• builds stakeholder confidence and trust
• clear, structured, and precise communicator with technical peers
• data-grounded, practical decision-making
• aligns teams and helps stakeholders make informed decisions