Release Notes
v1.6 — Semantic Search
Semantic search joins keyword search in a unified hybrid pipeline. Embeddings on every document, Qdrant Cloud as the vector store.
v1.6 — Semantic Search
Released 2025-08-19.
Added
- Semantic search — every uploaded document is embedded; queries are embedded and matched
- Hybrid ranking — combines SQL Server full-text score with semantic similarity
- Q&A endpoint at
/api/v1/search/qa— retrieval-augmented question answering against the tenant's corpus - Related documents panel on document detail view — top-3 semantically similar documents
Changed
- Search bar UI — now shows result type (Document / FAQ / Person / Contract) inline
- Search results — display semantic confidence score for transparency
Performance
- Embedding generation moved to background queue; document upload no longer blocks on embedding
- Vector store is Qdrant Cloud (per-tenant collections); average query latency under 80ms
Fixed
- Cross-language semantic search (English query → Swahili document) was producing biased results favouring English; now corrected with multilingual embedding model