Two problems:
(1) The AI doesn’t extract deadlines from emails
(2) embedEmail() and searchSimilarEmails() exist in code but have NO API endpoints or UI integration. We need to fix both.
Acceptance Criteria:
• Update AI service: add deadlines: string[] to structured output schema
• Add deadline field to ActionItem model, deadlines to EmailAnalysis model
• Create backend/src/routes/rag.routes.ts:
– POST /api/rag/index — trigger embedding for email(s)
– POST /api/rag/search — semantic search with cosine similarity
• Ensure pgvector extension enabled in PostgreSQL
• Add embedding vector(1536) to Email model (raw SQL for vector type)
• Create BullMQ background job indexEmailsJob for batch indexing
• Add semantic search UI to frontend (search bar with “AI-powered” toggle)
Technical Notes
• Use chrono-node as fallback if LLM misses dates
• Search must complete in <500ms for 100K+ emails
• Filter by user_id for privacy
• Must handle emails with no body (skip indexing)
Two problems:
(1) The AI doesn’t extract deadlines from emails
(2) embedEmail() and searchSimilarEmails() exist in code but have NO API endpoints or UI integration. We need to fix both.
Acceptance Criteria:
• Update AI service: add deadlines: string[] to structured output schema
• Add deadline field to ActionItem model, deadlines to EmailAnalysis model
• Create backend/src/routes/rag.routes.ts:
– POST /api/rag/index — trigger embedding for email(s)
– POST /api/rag/search — semantic search with cosine similarity
• Ensure pgvector extension enabled in PostgreSQL
• Add embedding vector(1536) to Email model (raw SQL for vector type)
• Create BullMQ background job indexEmailsJob for batch indexing
• Add semantic search UI to frontend (search bar with “AI-powered” toggle)
Technical Notes
• Use chrono-node as fallback if LLM misses dates
• Search must complete in <500ms for 100K+ emails
• Filter by user_id for privacy
• Must handle emails with no body (skip indexing)