ML Engineer — RAG Pipelines

Quick facts

About the role

Build the RAG layer for our enterprise search product. You'll own the embedding pipeline, the retrieval engine, the re-ranker, and (importantly) the eval harness that proves any change is actually an improvement. We're well past the demo-RAG stage — looking for someone who's debugged real-world retrieval failures and has opinions on chunk sizing, hybrid search, and where to spend the latency budget.

What you’ll do

• Own the embedding pipeline + retrieval engine end-to-end
• Define the eval harness and the regression test set
• Tune chunk sizing, re-ranking, and hybrid-search weights for the product domain
• Partner with product on the user-feedback loop and the failure-mode roadmap
• Mentor one junior ML engineer through their first RAG project

What we’re looking for

• Shipped a RAG pipeline that survived contact with real users
• Comfortable across multiple vector stores (Pinecone, Weaviate, pgvector)
• Has built an eval harness with both automated metrics + human review
• Strong Python; fluent in PyTorch when fine-tuning is on the table
• Knows where to push back when the right answer is "do not use an LLM here"

About Lucent Search

Lucent Search is a enterprise search company (11–50 people). Founded in 2023. Based in New York, NY. They hire and pay through Loxala, so scope, milestones, and funds stay protected by escrow from the first message.

Skills required