This project enables users to upload various document types (PDF, PPT, DOCX, CSV) and chat interactively with the content through a conversational interface. Powered by advanced embeddings and a state-of-the-art language model, it provides a secure and personalized experience for each user.
Watch Demo Video Showcasing Chabot in action
- Backend: Python, FastAPI, LangChain, ChromaDB, NVIDIA Embedding, Unstructured.io, Redis, Docker
- LLM: GROQ Llama 3.1 70b
- Frontend: ReactJS
- Authentication: JWT, OAuth2.0
- Storage & Session Management: Redis
- Containerization: Docker
- Navigate to the
serverdirectory:cd server - Deploy the backend using Docker:
bash deploy.sh
- Navigate to the
frontenddirectory:cd frontend - Deploy the frontend:
bash deploy.sh
-
User Authentication & Authorization:
- Registration and login with token-based session management.
- JWT tokens stored securely in Redis for session validation.
- Passwords are hashed using
bcryptand stored securely. - Only authenticated users have access to chat and document-related services.
-
Document Upload & Embedding:
- Users upload various document types (PPTX, PDF, CSV, DOCX).
- Document embeddings are generated with NVIDIA Embeddings via Unstructured.io.
- User-specific embeddings are saved in ChromaDB for private access.
- Documents are deleted post-processing for enhanced privacy.
-
Chat Interface with Retrieval-Augmented Generation (RAG):
- Users can submit queries related to document contents or general questions.
- The chatbot dynamically determines query context and retrieves responses.
- All queries and responses are stored in Redis to maintain conversation context.
-
Chat History Retrieval:
- Users can view the latest chat history for a continuous and personalized conversation experience.
-
Security & Data Privacy:
- Embeddings and user data are stored in user-specific collections for isolation.
- Redis stores only essential session and chat data, ensuring user privacy.
- Attributes:
user_id,email,password_hash,session_token - Functions:
register_user(email, password): Registers a new user.login_user(email, password): Authenticates and logs in the user.verify_token(session_token): Verifies the JWT for session continuity.
- Attributes:
file_path,user_id - Functions:
upload_document(file_path): Uploads user documents.process_and_embed_document(file_path, user_id): Generates embeddings for the document.delete_document(file_path): Deletes document after embedding.
- Attributes:
user_id,query - Functions:
send_message(query, user_id): Sends a query message.generate_response(query): Retrieves or generates a response using LLM.store_message(user_id, message, sender): Stores each message in Redis for history.
- Attributes:
user_id,session_token - Functions:
generate_jwt(user_id): Generates a JWT token for user sessions.verify_jwt(session_token): Validates the session token.logout_user(user_id): Logs the user out and removes the session.
- User interacts with SessionManager for secure authentication.
- User uploads documents via DocumentProcessor, which embeds the documents in ChromaDB.
- User submits queries to ChatAgent, which uses Redis to manage chat history and generate responses using embeddings and the LLM.
- ChatAgent is designed for extensibility, allowing the integration of new LLM models without altering its core structure.
- DocumentProcessor can support additional document types by implementing interface methods, enabling easy extension for future requirements.
To maintain modularity and flexibility:
- SessionManager and ChatAgent interact via interfaces, facilitating easy adjustments in session management.
- DocumentProcessor integrates with Unstructured.io, which can be abstracted to support other embedding tools as needed.
