🧠 Statistical Test Recommender Agent (LangGraph + SciPy)

This project is an AI agent that helps you choose the most appropriate statistical test based on a natural language question. Built using LangGraph, LangChain, and a locally embedded version of the scipy.stats documentation, the agent even returns ready-to-run Python code.

---

🚀 Project Description

"Which test should I use?"
Choosing the right statistical test can be confusing. This AI assistant breaks down your question, retrieves documentation if needed, asks clarifying questions, and recommends the correct test — plus generates the code using scipy.stats.

Streamlit App

Access the deployed App here:
https://ai-statistical-advisor.streamlit.app/

---

📦 Requirements

• Python 3.10+
• OpenAI API Key
• chromadb >= 1.0.16
• langchain-chroma >= 0.2.5
• langchain-community >= 0.3.27
• langchain-openai >= 0.3.28
• langchain >= 0.3.27
• langgraph >= 0.6.4
• openai >= 1.99.3
• streamlit >= 1.48.0
• tiktoken >= 0.10.0

---

▶️ How to Run

Streamlit

1. Access the App on http...
2. Get an OpenAI API key and add that to the sidebar on the app.
3. Write a question about statistical tests
4. Hit Search.

Locally

1. Clone this repo

git clone https://github.com/gurezende/AI-Statistical-Advisor.git

2. Run the app

stramlit run main.py

🧠 Code Structure

stat-test-advisor/
├── docs/ # Raw documentation (for RAG)
├── langgraph_agent/
│ ├── graph.py # LangGraph build logic
│ ├── nodes.py # All nodes (classify, clarify, retrieve, respond, code)
├── rag/
│ ├── embedder.py # Chunk + embed docs
│ ├── retriever.py # Load and query embedded docs
├── main.py # CLI interface to run the agent
├── embed.py # One-time embed script
├── requirements.txt

🤖 Agent Logic (LangGraph)

graph TD
    A[Start: User Question] --> B[classify_intent]
    B -->|search| C[retrieve_info]
    B -->|simple| D[respond]
    B -->|clarify| E[ask_clarification]
    E --> B
    C --> D
    D --> F[generate_code]
    F --> G[End]

Loading

💡 Demonstration

Sample Input:

I want to compare the means of two independent groups.

Output:

👨‍🔬 About

Created by Gustavo R Santos as a learning and teaching project using:

• LangGraph for agentic reasoning
• LangChain for RAG and embeddings
• SciPy for statistical tests

📄 License

Project licensed under MIt License.