CVBot Backend

A sophisticated AI-powered personal assistant backend designed to provide seamless access to your professional profile through intelligent document retrieval, context-aware responses, and proactive suggestions.

🚀 Features

• Intelligent Agent (LangGraph)
  Retrieves relevant information from your documents and alerts you when knowledge gaps are detected.
• Suggestions Engine
  Generates follow-up or suggested questions to guide more effective interactions.
• Document Processing
  Advanced OCR and text pipeline supporting multiple file formats with smart chunking.
• Vector Search
  Efficient semantic retrieval with ChromaDB.
• Notifications
  Automatic email alerts sent via SMTP when missing or incomplete information is found.
• File Management
  Full CRUD support for documents in the vector database.
• API Tracing & Monitoring
  Powered by Opik for observability and debugging.

🛠️ Tech Stack

• Framework: FastAPI
• Agent: LangGraph
• Database: PostgreSQL + ChromaDB (vector storage)
• Cache / Rate Limiting: Redis
• Authentication: Clerk
• OCR: Docling + EasyOCR
• Monitoring: Opik
• Language: Python
• Email Notifications: SMTP

📋 Architecture

Document Processing Pipeline

1. File Upload – PDF files processed with Docling for OCR extraction.
2. Markdown Conversion – Unified content representation.
3. Smart Chunking – Content split by headings and intelligently chunked with an LLM.
4. Context Enhancement – Each chunk enriched with surrounding context.
5. Vector Embedding – Enhanced chunks stored in ChromaDB for semantic retrieval.

Agent Workflow

• Query Processing – User queries handled by the LangGraph agent.
• Vector Search – Relevant embeddings retrieved from ChromaDB.
• Response Generation – Professional, contextually accurate responses produced.
• Suggestions Node – Asynchronous node that generates question suggestions to improve user interaction.
• Notification System – Missing information triggers email notifications via SMTP.

flowchart TD
    U[User Query] --> Q[LangGraph Agent]
    Q --> R[Vector Search in ChromaDB]
    R --> S[Response Generation]
    S --> T[User Response]
    Q --> SG[Suggestions Node]
    SG --> T
    Q --> N[Missing Info Detected]
    N --> EML[Email Notification via SMTP]

Loading

🛤️ Project Routes

The CVBot Backend exposes modular routes under /api/v1:

• /vector-store – Manages documents in the vector DB.

  • POST /vector-store/store-files: Upload & process documents.
  • POST /vector-store/semantic-search: Perform semantic search.
  • GET /vector-store/retrieve-files: List processed files.
  • GET /vector-store/retrieve-embeddings?filename=...: Retrieve embeddings for a file.
  • DELETE /vector-store/delete-files: Remove files.

• /projects – Manage portfolio projects.

  • GET /projects: List all projects.
  • POST /projects: Add new project.
  • DELETE /projects/{project_id}: Delete project by ID.

• /chatbot – AI chatbot interaction.

  • POST /chatbot/invoke: Send a message to the AI agent. (Rate limited: 20 req/min per user)
  • GET /chatbot/history?session_id=...: Retrieve session history.

---

vector-store and projects routes are secured with Clerk authentication.

---

📊 Monitoring

• Tracing & monitoring with Opik for reliable performance and debugging.

🔍 Document Processing Details

Supported File Formats

• PDF: Processed with Docling (OCR support).
• Markdown: Directly processed.

Chunking Strategy (Inspired by Anthropic's Contextual Retrieval)

1. Heading-based Split – Structural splitting.
2. LLM-powered Chunking – Optimized chunk sizes.
3. Context Addition – Enriched with surrounding text.
4. Semantic Embedding – Converted to vectors for similarity search.

🤖 Agent Capabilities

The LangGraph-powered agent provides:

• Semantic search through professional documents.
• Contextual answers about experience & skills.
• Identification of knowledge gaps with SMTP email alerts.
• Conversation context maintenance for seamless dialogue.
• Professional, tailored responses.
• Suggested follow-up questions for improved engagement.

🔐 Security

• Authentication: Clerk-based authentication.
• API Security: Redis-based rate limiting & strict input validation.

💻 Installation

1. Navigate to the deployment folder.
2. Copy .env.example → .env and configure environment variables (incl. Redis & SMTP).
3. Run docker compose up to start services (includes Redis).
4. Containers initialize; NGINX exposes port 80 → host 8081 (configurable).
5. Access via localhost:8081 or a reverse proxy.

👨‍💻 Author

Originally built by Kaloyan Stefanov.