Chen Wei
ML engineer. Retrieval, evals, and small-team production ML.
About
ML engineer specializing in retrieval systems and eval frameworks. Previously built the search relevance stack at a mid-size marketplace; currently applying the same eval-first mindset to LLM products at a seed-stage startup.
Selected projects
eval-lite
1,400A minimal eval framework for LLM applications — writes results to Postgres, plots regressions across runs, no vendor lock-in.
hybrid-retriever
720A reference implementation of hybrid retrieval — BM25 + dense + reranker — that outperforms single-mode on a bench of six datasets.
Writing
Interview prep
ML systemsHow do you evaluate a RAG pipeline?
Two axes: retrieval (recall@k, MRR — did we fetch the right chunks) and generation (faithfulness, answer relevance — did we use them correctly). Evaluate the two separately or you can't diagnose regressions.
Resume tips
For ML roles, one line per model you shipped: the objective, the baseline you beat, and the metric it moved in prod. `Retrieval CTR +8.1% vs BM25 baseline` beats any list of frameworks.