History of Nepal Chatbot

Overview

The History of Nepal Chatbot is an intelligent conversational agent designed to provide users with accurate, in-depth, and engaging information about the history of Nepal. It integrates multiple advanced technologies and tools to ensure high-quality responses, leveraging retrieval-augmented generation (RAG) techniques, knowledge graphs, and internet-based searches.

---

Features

Core Functionalities

• Knowledge Sources:
  • Qdrant: For dense retrieval using embeddings.
  • Knowledge Graph (Neo4j): To explore relationships among historical entities.
  • Google Search: To supplement responses with up-to-date information.
• Conversation Handling:
  • Session-based interactions using LangChain.
  • Stores old conversations for context-aware responses.
• Custom Agent:
  • Orchestrates workflows to retrieve and synthesize data seamlessly.
• Data Persistence:
  • Query and response storage in PostgreSQL with session IDs and timestamps.

Infrastructure

• Fully Dockerized architecture for ease of deployment.
• Integration of multiple services:
  • Ollama for language modeling.
  • LangChain for query handling.
  • PostgreSQL for database storage.
  • Neo4j for graph-based data representation.
  • Qdrant for vector-based search.

Additional Features

• Multi-source validation:
  • Google Search validates outputs from Qdrant and Neo4j.
  • Replaces inconsistent or outdated information with Google search results.
• Session Management:
  • Tracks and retains user sessions for enhanced interaction.
• Time-Stamped Logs:
  • Maintains a history of interactions for user reviews and analytics.

---

Technology Stack

Backend

• Ollama: Language model for generating human-like responses.
• LangChain: Manages interaction workflows and context handling.

Data Management

• Qdrant: Vector database for document retrieval.
• Neo4j: Graph database for entity relationships.
• PostgreSQL: Stores query-response pairs with session data.

Deployment

• Docker: Containers for all components, ensuring portability and scalability.

External Services

• Google Search API: Provides supplementary information and validation.

---

Installation and Setup

Prerequisites

• Docker
• Docker Compose

Steps

1. Clone the repository:

   git clone <repository_url>
   cd <repository_directory>

2. Create a .env file for environment variables. Example:

   • Take reference from env_description which will be required to run this entire project,

3. Build and start the Docker containers:

   docker-compose up --build

4. Access the chatbot service via the designated API endpoint or UI (if available).

---

Run the program

• Step 1: Parse the document. Use preprocessing/parse.py which will parse the document using llama parse and store it in markdown format
• Step 2: Then to extract the document and chunk them use cmd/extraction_script.py which will chunk the parsed document.
• Step 3: To push to the qdrant database and perform embedding and perform hybrid search use cmd/hybrid_search_script.py
• Step 4: To push to the neo4j database and perform embedding and perform retrieveral from graph database use cmd/ neo4j_graph_script.py
• Step 5: To run agent use agents/custom_agent.py

Technologies used

• Llama3.2 3B
• langchain agent
• neo4j
• embedding model (mxbai-embed-large)
• sparse embedding (Qdrant/bm25)
• NLLB 600M from ctranslate2
• rerank cohere

Usage

1. Ask Historical Questions:

   • Example: "What is the significance of the Battle of Kirtipur?"
   • The chatbot will retrieve information from Qdrant, Neo4j, and validate with Google Search.

2. Explore Relationships:

   • Example: "What was the relationship between Prithvi Narayan Shah and the British East India Company?"

3. Session Continuity:

   • Engage in conversations where the chatbot retains the context of previous queries.

---

API Endpoints

1. Chat

• Endpoint: /api/chat
• Method: POST
• Payload:

  {
    "query": "What is the history of Lumbini?",
    "session_id": "12345"
  }

2. Retrieve Old Conversations

• Endpoint: /api/conversations
• Method: GET
• Query Parameters:
  • session_id: Session identifier

---

System Architecture

1. User Query:

   • Input query received via API or UI.

2. Retrieval:

   • Query sent to Qdrant for vector-based search.
   • Query sent to Neo4j for graph traversal.

3. Validation and Augmentation:

   • Google Search API validates and supplements retrieved data.

4. Response Generation:

   • Data synthesized using Ollama and LangChain.
   • Response stored in PostgreSQL with session ID and timestamp.

5. Final Output:

   • Context-aware and validated response delivered to the user.

---

Future Enhancements

1. Interactive Visualizations:
   • Timelines, maps, and graphs for richer user experiences.
2. Voice Integration:
   • Enable voice-based queries and responses.
3. AI-Driven Insights:
   • Predictive analytics and insights about historical trends.

---

Contributing

Contributions are welcome! Please follow the steps below:

1. Fork the repository.
2. Create a new branch:

   git checkout -b feature-branch-name

3. Commit your changes and open a pull request.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Acknowledgements

• The creators of Ollama, LangChain, Qdrant, Neo4j, and PostgreSQL for providing robust tools.
• Google Search API for real-time supplementary information.

---

For any issues or queries, please contact [[email protected]].