Saikrishna Paila

About

Built and shipped production RAG/agentic systems in high-stakes domains (biomedical and legal), including an enterprise biomedical document retrieval platform over ~10k scientific docs and a multilingual African-law assistant at the World Bank. Deep hands-on experience with LangChain/LangGraph/LlamaIndex and evaluation tooling (LLM-as-a-judge, safety/hallucination detection), with measurable gains in retrieval quality and hallucination reduction.

Key Strengths

• Built and deployed production AI access system for biomedical document retrieval (10,000 scientific documents) for enterprise customers
• Improved retrieval rate by ~40% using hybrid search (dense embeddings + BM25) with cross-encoder re-ranking
• Reduced hallucinations ~40% via LangGraph stateful, multi-step workflows with validation/refinement loops
• Designed LLM-as-a-judge evaluation with multidimensional scoring (factuality/groundedness/completeness/safety) and JSON-structured outputs
• Improved judge reliability by running multiple judge instances for cross-verification and using structured prompts
• Reliability-first agent design using explicit failure-mode mapping (retrieval misses, hallucinations, tool failure, infinite loops) with mitigations
• Strong testing discipline across unit/integration/adversarial/regression plus human eval and A/B testing
• Effective collaboration with non-technical stakeholders using demos, regular check-ins, decision logs, and explicit trade-off discussions

Experience

Education

Awards

• 2nd Place in the AI Track sponsored by Meta

Skills

Languages

Certifications