E-005: Document AI & RAG Context

Status: In Progress Owner: @bilal Priority: P1 — Ship to Production

Objective

Enable the AI to reference property-specific documents (house rules, boiler manuals, gas safety certs) when responding to tenants. Complete the document AI pipeline from upload to RAG retrieval.

What’s Done

• visibleToAI toggle exposed in GraphQL + document modal UI
• AI indicator badge on document cards
• Seed script with house rules + boiler manual (2 docs, 3 chunks, stub embeddings)
• pnpm db:seed:documents command
• Property context included in system prompt (address, type, units, room)

What’s Next

Task	ID	Description	Status
Text extraction pipeline	DOC-AI-003	On upload with visible_to_ai = true, extract text (PDF/image OCR), chunk, embed into document_chunks	Next
Test with real docs	CTX-002	Upload real property documents, verify RAG retrieval quality	Blocked by DOC-AI-003
Document suggestions	CTX-003	AI recommends what landlord should upload based on property type	Planned
AI configurability	CTX-004	Per-property/org AI behaviour settings (tone, topics, DIY policy)	Planned

Key Decisions

• Extraction: PDF text extraction + image OCR (for scanned documents)
• Chunking: 300-500 tokens, semantic paragraph-level, sentence-boundary overlap (see ADR-018 RAG Context Management)
• Embedding: OpenAI text-embedding-3-small (1536 dimensions)
• Storage: document_chunks table with pgvector index
• What AI knows (for triage) vs what it tells the tenant — configurable per org

Dependencies

• Supabase Storage (done — documents uploadable)
• pgvector extension (done — enabled locally)

Related

• ADR-017 RAG Pipeline v2
• ADR-018 RAG Context Management
• E-010 Retrieval Improvements (hybrid search, depends on this epic)