At Elastic, I designed an access-aware RAG platform across 2.4 million enterprise documents, raising answer acceptance from 61% to 86% with Elasticsearch search and citation grounding.
I engineered governed LangGraph workflows with approved tools, permission controls, human approvals, and audited execution. These workflows automated 72% of routine case-preparation activities.
To improve AI quality, I built a 1,500-case regression framework covering retrieval, groundedness, citations, policy compliance, prompt injection, and tool execution. It reduced critical production AI defects by 54%.
I also deployed Python/FastAPI services on Kubernetes and instrumented them with OpenTelemetry and Elastic Observability. This work reduced p95 response time to 2.1 seconds and cost per successful task by 37%.
