SmartDoc LangGraph Agent

Document-agent prototype for PDF question answering, calculator-assisted workflows, and dynamic tool routing with LangGraph.

The repository demonstrates an enterprise-relevant agent pattern: route user requests to the right tool, retrieve evidence from uploaded documents, answer with document context when available, and fall back to a controlled utility tool for arithmetic.

What It Demonstrates

• Document question answering: PDF retrieval through a LangChain vector store
• Tool routing: LangGraph workflow for selecting between document QA, calculator, and general LLM paths
• Local model path: TinyLlama-compatible setup for CPU-friendly experimentation
• Reviewer UI: Streamlit interface for testing document and tool-routing behavior
• Extensible agent design: Small codebase that can be replaced with stronger models, rerankers, or enterprise document stores

Scope

This is a compact agent prototype, not a production document platform. A production version would add authentication, document-level access control, citation enforcement, structured logging, evaluation sets, and model/embedding observability.

---

Features

• PDF document question answering
• Calculator for simple math queries
• General-purpose LLM question answering
• Dynamic routing via LangGraph
• Streamlit UI for local review
• Hugging Face Transformers compatibility
• CPU-friendly local execution path

---

Requirements

• Python 3.10+
• Anaconda (recommended)
• git
• Hugging Face Token (required to download TinyLlama)

---

Installation

# Clone this repo
https://github.com/rajantripathi/smartdoc-langgraph-agent.git
cd smartdoc-langgraph-agent

# Create virtual environment (Recommended)
conda env create -f environment.yml
conda activate agentenv

# Or install manually
pip install -r requirements.txt

---

Usage

1️⃣ CLI Version

python run_langgraph.py

2️⃣ Streamlit UI

streamlit run streamlit_app.py

---

Folder Structure

smartdoc-langgraph-agent/
├── agent.py
├── langgraph_agent.py
├── pdf_reader.py
├── pdf_qa_tool.py
├── tools.py
├── tool_selector.py
├── run_langgraph.py
├── streamlit_app.py
├── requirements.txt
├── environment.yml
├── data/
│   └── sample.pdf
└── .env.example

---

Notes

• The project demonstrates a small but clear LangGraph routing pattern.
• TinyLlama can be replaced with another Hugging Face-compatible model.
• The code is intentionally compact so the orchestration logic is easy to inspect.

---

Credits

Built by Rajan Tripathi with support from LangGraph, LangChain, and HuggingFace libraries.

---

License

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