This is exactly what Vespa is good for. It's a unique open source project combining
- Sparse (WAND/BM25)
- Dense (ANN/HNSW)
- Hybrid (WAND OR ANN)
- Multiphase re-ranking with many types of ML models (neural, xgboost,gbdt )
- Cost effective, Fast and scalable in any dimension
💯 This makes integrating neural net based semantic search with traditional BM25 practical in production.
Having a single project that integrates both solves the problems in trying to glue a vector search library or database with another document search.