docqube_backend/app/modules/chat

Chatbot Module

This module implements a robust, conversational RAG (Retrieval-Augmented Generation) chatbot for documents, integrated into the DocQube application.

📂 Structure

• engine/: The core RAG orchestration logic (using LlamaIndex and custom adapters).
  • chatbot_engine.py: Main orchestrator handling intent, memory, and path selection.
  • intent.py: Heuristic and LLM-based intent classification.
  • retrieval.py: Qdrant-based retrieval pipeline.
  • rewrite.py: Context-aware query rewriting.
  • prompts.py: Expertly engineered RAG and intent prompts.
  • formatter.py: Dynamic response formatting logic.
• services/: Integration services.
  • chat_service.py: FastAPI service layer managing session engines and streaming.
• models/: SQLAlchemy and Pydantic models for chat data.
• routes/: FastAPI endpoint definitions.

🏗️ Architecture

User Query → Intent Detection → Query Rewrite → Retrieval → LLM (DeepSeek) → Streaming Response

• LLM: DeepSeek Chat (via OpenAI-compatible adapter).
• Vector DB: Qdrant (integrated via LlamaIndex/Custom pipeline).
• Embeddings: BAAI/bge-base-en-v1.5 for semantic search.

⚙️ Configuration

• DeepSeek API: Requires DEEPSEEK_API_KEY in .env.
• Vector Store: Qdrant must be running (configured via docker-compose).
• Persistence: Engine indices are cached per-document for performance.

🚀 Endpoints

• POST /api/chat/upload: Upload a document for analysis.
• POST /api/chat/chat: Synchronous chat response.
• GET /api/chat/stream: Streaming chat response (Server-Sent Events).
• GET /api/chat/documents: List uploaded documents.
• DELETE /api/chat/documents/{id}: Delete a document and its indices.