Math Agent

This project implements a sophisticated Math Agent using Langflow. The agent is designed to solve mathematical problems by leveraging a combination of a local knowledge base, web search capabilities, and powerful language models.

Features

• Knowledge Base: Utilizes a Qdrant vector store to maintain a persistent knowledge base of math problems and their solutions.
• Web Search: Integrates with the Tavily API for real-time web searches when a solution is not found in the local knowledge base.
• Advanced Language Models: Powered by Google's Gemini Flash 2.0, capable of handling large input tokens for complex problems. It initially used Groq providers with Llama 4 Scout.
• Guardrails:
  • An input guardrail, using a system prompt, ensures the agent only responds to math-related queries.
  • A qwen3 4B model is used for output guardrails, leveraging its local tool-calling capabilities.
• Human-in-the-Loop: Includes a review process for users to validate and approve new solutions before they are added to the knowledge base.
• KB Management: A Node.js server with dedicated endpoints manages fetching and updating the knowledge base from a JSON file.
• Embeddings: Uses nomic-embed for generating high-quality embeddings for the Qdrant vector store.

Prerequisites

• Python 3.10
• Node.js and npm
• An instance of Qdrant (local or cloud)
• API keys for:
  • Google Gemini
  • Tavily

Installation and Setup

1. Clone the repository:

   git clone <repository-url>
   cd <repository-directory>

2. Install uv (a fast Python package installer):

   pip install uv

3. Create a virtual environment and install Python dependencies:

   uv venv
   source .venv/bin/activate  # On Windows use `.venv\Scripts\activate`
   uv pip install -U langflow 

4. Set up the Node.js server:

   npm install

5. Configure Environment Variables: Create a .env file in the root of the project and add your API keys:

   GEMINI_API_KEY="your_gemini_api_key"
   TAVILY_API_KEY="your_tavily_api_key"
   QDRANT_URL="your_qdrant_instance_url"
   QDRANT_API_KEY="your_qdrant_api_key"
   

How to Use

1. Start the Node.js server for KB management:

   node node_server.js

2. Run the Langflow application:

   uv run langflow run

3. Access the Langflow UI: Open your web browser and navigate to http://127.0.0.1:7860.

4. Import the Flow: Import the Math Agent.json file into the Langflow interface to load the pre-built agent.

5. Interact with the Agent: Use the chat interface in Langflow to ask the Math Agent your math-related questions.

Math Agent

Git Link for the Math Agent Frontend https://github.com/NandeeshaHK/quant-talk