AI Engineer in the making — I build systems that put LLMs into production, not just notebooks.
Computer Science @ UC Irvine · Focused on RAG systems, LLM evaluation, and applied ML infrastructure
I'm working toward AI Engineer / Forward Deployed Engineer roles — the kind of work where you're not just calling an LLM API, but designing the retrieval, evaluation, and production architecture around it so it's actually trustworthy at scale. Right now that means:
- 🔎 RAG systems — retrieval strategy (hybrid search, reranking, chunking design), not just "stuff it in a vector DB and hope"
- 📊 LLM evaluation — faithfulness, context precision, and failure analysis as first-class engineering work, not an afterthought
- ⚙️ Applied ML infrastructure — the boring-but-essential parts: ingestion pipelines, structured outputs, test coverage, reproducibility
- 🧪 Comfortable being wrong in public — I report what the data actually says, even when it contradicts my own hypothesis
An internal support RAG assistant over a 100-document enterprise knowledge base — hybrid (BM25 + vector) retrieval fused with RRF, structured/grounded generation with explicit confidence scoring, and a full RAGAs evaluation harness.
- 0.939 faithfulness · 0.775 context precision against a 50-query eval set (thresholds: 0.70 / 0.60)
- Measured hybrid search against vector-only as a real before/after comparison — and reported honestly when the data didn't confirm my hypothesis
- Full failure-case analysis: distinguished retrieval misses from generation failures from incomplete cross-document synthesis, with root causes for each
Core / ML
LLM / RAG stack
Data & infra
