🧠 Refract: An AI-Powered Research Agent

Refract is an intelligent, agentic AI assistant designed to support researchers, students, and knowledge-seekers in exploring complex topics.
Built using the ReAct (Reasoning + Acting) paradigm, it leverages large language models and real-time web tools to reason, search, and summarize information step-by-step — just like a human researcher.

---

🔍 Key Features

• 🤖 LLM-powered Agent using Qwen/Qwen1.5-1.8B-Chat
• 🧠 ReAct Loop: Think, act (search/scrape), and reflect iteratively
• 🌐 Web-Augmented: Uses Playwright for dynamic web search and scraping
• 🛠️ Tool-Use Capable: Executes Search[...] and Scrape[...] commands during reasoning
• 📄 Auto-Generated Reports in PDF and Markdown
• 🧪 Partial Summary Fallback: When the reasoning step limit is hit, Refract summarizes and verifies what it has collected so far
• 🧭 Extensible Design for integrating new tools and models

---

📦 Tech Stack

• transformers + torch for language model inference
• playwright + beautifulsoup4 for real-time web scraping
• langchain for tool abstraction and compatibility
• reportlab for structured PDF report generation

---

🚀 How It Works

1. User enters a topic (e.g., "Foundation Models in Machine Learning")
2. Refract reasons about the topic and decides when to take actions like Search[...] or Scrape[...]
3. It gathers evidence, reflects, and eventually produces a concise, verified summary
4. If it hits the max step limit, it summarizes what it found and runs a verification pass

---

📁 Folder Structure

Refract/
├── main.py # Main entry point for the agent
├── llm.py # Model logic and ReAct prompt templates
├── tools.py # Executes Search[...] and Scrape[...] actions
├── playwright_search_scrape.py # Search and scrape logic using Playwright
├── report_utils.py # Generates reports in Markdown and PDF formats
├── requirements.txt # Python dependencies
├── README.md

---

⚙️ Setup Instructions

1. Clone the repo

git clone https://github.com/yourusername/refract.git
cd refract

2. Install Python dependencies

pip install -r requirements.txt

3. Install Playwright browser drivers

playwright install

4. Run the agent

python main.py

---

✅ Requirements

• Python 3.8+
• Internet connection for model and scraping
• GPU (recommended) for faster model inference

---

✨ Future Improvements

• Create own ReAct prompt
• Support for more tools (e.g., Wolfram, Wikipedia API, Paper API)
• RAG-based memory integration
• Support for larger Qwen variants or open-source alternatives
• Web UI using Gradio or Streamlit

---

📜 License

This project is licensed under the MIT License

---

👨‍💻 Author

Karthik B Magadi
GitHub: @Magnus0969

---

⭐ Star this repo if you found it helpful!