On-demand Faiss vector search from S3. Lazy-loads clusters without fetching entire index. Original implementation: https://github.com/at15/faiss/tree/at15/faiss-on-s3/demos/s3_ivf
- e2e test script in python
- github action (with cache on faiss and aws-sdk?)
- benchmark using larger dataset
- Add how to build server
python3.13 -m venv .venv
source .venv/bin/activate
# Brings in boto3 and faiss-cpu
pip install -e ./client
# Generate embeddings
pip install sentence-transformers pandasMIT