Search & documents

Hybrid search and grounded Q&A over your uploaded and SharePoint-synced documents — semantic, keyword and filename ranking fused, with access control baked into the index.

Book a walkthrough Back to overview

See it

ExampleGrounded answer with cited source documents

ExampleFaceted document search across a project

What it does

Tri-modal retrieval

Reciprocal Rank Fusion of semantic (embeddings), keyword and filename rankers, with mining-domain phrase boosts.

Grounded answers with citations

Docs-only answers from your documents, with source citations — no ungrounded summaries.

SharePoint sync

Microsoft Graph sync of files into the index, with text extraction and per-user site scoping.

Real-time index

Whoosh/Haystack full-text with index-on-save; separate document and SharePoint indexes.

Permission-aware

Access-level, client/project and finance-permission facets baked into the index — retrieval respects gating.

Cleaner context

Per-source caps, neighbour-chunk expansion and spreadsheet down-weighting produce tighter context.

How it works — the engineeringTechnical detail

Google Gemini embeddings plus Gemini answering with context caching; Whoosh/Haystack with a real-time signal processor; a chunked store; SharePoint via the Graph API.

core/ai_hybrid_search.py · ai_embedder.py · views_ai.py · search_indexes.py