Rui Zhao

About

Built and shipped a production RAG system (“TROJAN KNOWLEDGE”) for answering questions over technical PDFs, using a 3-stage retrieval stack (BM25 + FAISS + cross-encoder) to lift F1 from 71% to 84%. Drove major performance gains with a 3-level cache (memory/Redis/disk) cutting latency from ~200ms to ~10ms, and added Prometheus/Grafana monitoring plus LangChain-based fallback logic to handle OpenAI rate limits under load.

Experience

Education

Key Strengths

• Built and deployed production RAG system for technical PDF Q&A
• Improved retrieval performance from 71% to 84% F1 via 3-stage retrieval (BM25 + FAISS + cross-encoder re-rank)
• Designed multi-level caching (memory/Redis/disk) reducing latency ~200ms to ~10ms (~20x+ faster)
• Implemented reliability strategies for LLM rate limits (caching + fallback to smaller context windows)
• Set up observability with Prometheus/Grafana and focuses on latency metrics like P95
• Established automated evaluation in CI/CD with benchmark-based regression detection

Reference Highlights

• Highly helpful technical mentor/consultant
• Provided clear end-to-end guidance for building a RAG prototype
• Articulate in breaking down stakeholder needs
• Proactive at identifying gaps and edge cases
• Able to adapt solutions to nuanced healthcare processes
• Strong at explaining fundamentals and big-picture concepts
• Patient and reassuring when others don’t understand
• Contributed to a prototype that worked well