README.mdView résumé
AI / ML · RETRIEVAL · MLOPS

Search systems that work the way people actually write.

I build retrieval and machine-learning systems together with the pipelines that train, evaluate, serve, and monitor them. My strongest work joins careful baselines with interfaces people can run.

0.983Recall@10 · Roman-Urdu retrievalfrom 0.192 on the same split
189 msMillion-row filter · Rustlocal, without uploading data
98.33%News classification0.9830 macro-F1 held out
5 stagesML lifecycle in one repoingest to monitoring

Pinned repositories

Loading public work…
Currently building active

Raabta

Cross-script retrieval for Roman-Urdu questions over Urdu-script documents, with three retrieval routes, reranking, and explicit abstention.

Inspect source