🧠 Advanced RAG Playground — BM25 + Embeddings + Re-ranking
🧠 Advanced RAG Playground — BM25 + Embeddings + Re-ranking
Launcher for the standalone AI: hybrid retrieval over documents you paste in — BM25, hashed dense vectors, HyDE, reciprocal rank fusion and MMR — with per-pass scores and a trace.
The RAG playground became the Knowledge mode of the standalone AI: index your own text locally, then search it with BM25 + dense + HyDE fused by RRF and diversified by MMR, with every intermediate ranking visible in the trace.
Open in the standalone AIOpens at: Knowledge mode → index + retrieval · runs entirely in your browser · no account, no API key, nothing uploaded
What the standalone AI gives you for this
- BM25 (k1=1.2, b=0.75), 64-dim hashed dense vectors and HyDE hypothetical answers
- Reciprocal rank fusion and tunable MMR lambda
- Search results feed every reasoning method, cited as [source]
The interactive tool that used to live on this card now runs inside the standalone AI at ai.html, where it shares your indexed documents, stored memories and every reasoning method. This card stays in the catalogue as a launcher so existing links keep working.