🧠 RAG Agent with Calculator + Knowledge Base

Powered by LangGraph + Claude

This is my first Retrieval-Augmented Generation (RAG) project that combines:

• 📚 Semantic search over documents
• 🧮 A calculator tool for precise math
• 🔗 Agent workflows using LangGraph
• 🤖 Claude for reasoning and response generation

---

🚀 Features

• 🔍 Knowledge Base Search

  • Uses FAISS for fast vector similarity search
  • Supports querying custom documents

• 🧮 Calculator Tool

  • Handles numerical queries accurately
  • Avoids LLM hallucinations in math

• 🔄 LangGraph Agent Workflow

  • Routes queries intelligently between tools
  • Supports multi-step reasoning

• 🤖 Claude Integration

  • Generates context-aware, high-quality responses

---

🏗️ Project Structure

.
├── faiss_index/        # Stored vector database
├── sample_docs/        # Input documents for RAG
├── agent.py            # LangGraph agent logic
├── ingest.py           # Document processing & embedding
├── app.py              # Main app entry point
├── main.py             # Alternate runner / testing
├── experiments.ipynb   # Experiments & testing
├── .env                # API keys
├── requirements.txt / pyproject.toml
└── README.md

---

⚙️ Setup

1. Clone the repo

git clone https://github.com/your-username/rag-agent.git
cd rag-agent

2. Create virtual environment

python -m venv .venv
.venv\Scripts\activate   # Windows

3. Install dependencies

pip install -r requirements.txt

---

🔑 Environment Variables

Create a .env file:

ANTHROPIC_API_KEY=your_claude_api_key

(Optional depending on embeddings)

OPENAI_API_KEY=
GOOGLE_API_KEY=

---

📥 Ingest Documents

python ingest.py

This will:

• Load documents from sample_docs/
• Convert them into embeddings
• Store them in faiss_index/

---

▶️ Run the App

python app.py

---

🧪 Example Queries

• “Summarize the documents”
• “What is discussed in the knowledge base?”
• “What is 125 * 48?”
• “Find and calculate key metrics from the data”

---

🧠 How It Works

1. User query is passed to the LangGraph agent

2. Agent decides:

   • Use calculator tool OR
   • Perform vector search

3. Relevant context is retrieved from FAISS

4. Claude generates the final response

---

🔥 Key Learnings

• Built a full RAG pipeline from scratch
• Integrated tools into an agent workflow
• Learned vector databases (FAISS)
• Handled real-world issues like rate limits