Krishna Kandlakunta

About

Built and deployed a production compliance automation RAG system (at Citi) that generates citation-backed, schema-validated risk summaries for regulatory document review. Emphasizes regulated-environment reliability with retrieval-only grounding, abstention, confidence thresholds, and immutable audit logging, plus orchestration using LangChain/LangGraph and Airflow. Reported ~60% reduction in compliance review effort while maintaining high precision and traceability.

Experience

Education

Key Strengths

• Built and deployed a compliance-focused RAG platform automating regulatory document reviews
• Strong hallucination mitigation via retrieval-only grounding, abstention behavior, and mandatory citations
• Auditability in regulated environments through full request logging and immutable audit trails (prompt/model versioning, retrieval metadata)
• Production optimization for latency and cost using model routing and caching
• Designs reliable agent/workflow systems with structured outputs, scoped tool access, retries, and guardrails
• Rigorous evaluation approach using gold datasets, regression/edge-case testing, A/B tests, and canary releases
• Defined and tracked grounded answer rate; improved quality by tightening retrieval thresholds and enforcing citation-backed outputs
• Delivered measurable impact: ~60% reduction in compliance review effort/time with improved consistency and traceability
• Effective collaboration with non-technical compliance/risk stakeholders by translating requirements into measurable metrics and HITL controls