AI & DATA SYSTEMS · 2025 to Present
CareSight
Personal projectMaking clinical information searchable.
Data preparationPartitioned processing connects the clinical-note corpus to embedding and indexing workflows.
Retrieval accessFastAPI exposes FAISS-backed searches over the processed note corpus.
THE PROBLEM
What needed solving.
A large collection of clinical notes needs a searchable representation that an application can retrieve through an API.
What I implemented.
Retrieval & data pipeline development
- Implemented FAISS vector search behind FastAPI endpoints as part of a retrieval-augmented generation pipeline for clinical-note analysis.
- Built partitioned Airflow ETL workflows to prepare clinical-note datasets for retrieval and downstream analytics.
- Connected data preparation, vector indexing, and retrieval API workflows so the application could search the processed note corpus.
ENGINEERING DECISIONS
The technical challenge.
The technical challenge was connecting data preparation to useful retrieval. Partitioned Airflow processing prepares the corpus, FAISS provides vector search, and FastAPI exposes retrieval endpoints. Data preparation and request handling remain separate parts of the workflow.
THE RESULT
What changed.
Combined partitioned data processing, vector search, and API access in a clinical-note retrieval workflow.
Ask me about the project