Send actual images to Gemini for multimodal RAG

Finished codebase

Fix: create conversation on upload for doc scoping

Fix: import Form from fastapi

Per-conversation document scoping for retrieval

Fix retrieval: two-stage doc-aware search

Fix email validation, re-index on delete, lazy-load DocumentIndex

Add Groq fallback for rate limits

Improve AI prompt to be conversational and helpful

Fix OOM: CPU-only PyTorch + fully lazy imports

Fix OOM: lazy-load ML models, rename to VectorMind

Add Docker and HF Spaces config

Add readme

feat: multi-conversation chat system with AI-generated titles

feat: streaming responses, canvas UI, audio upload, file management, clear data

feat: add frontend and backend code for multimodal RAG knowledge assistant