RAG with retrieval evaluation
Retrieval evaluation is the gate. The generator comes after the retriever earns its score.
- nDCG@10
- 0.137
- Baseline
- BM25
Work index
Six featured projects ranked by hiring signal, and an archive kept honest. Every case study states the result, the baseline, the evaluation design, and what is still missing.
Filter by track
Showing 10 of 10 projects
Retrieval evaluation is the gate. The generator comes after the retriever earns its score.
An ECG image goes in. A digitized signal, rhythm, axis, PR, QRS, and a review flag come out.
Five models, out of fold predictions, and a baseline table that keeps the stack honest.
From warehouse SQL to a Power BI view a team can act on.
Latency is the product. Tool calls are the risk.
Elo is the baseline, not the finish line.
Useful breadth, kept honest. These stay below the featured six because the hiring signal is secondary, not because the work is unfinished.
Two small projects, kept in the archive where they belong.
The lab that proves the security tools, not the AI story.
Lab work, evidence in the case study
Client sites built to ship, not to demo.
The site you are reading, built with the same rules as the case studies.