AGI Agent Playground

A modular, multi-agent pipeline for document understanding and reasoning using only free, local Hugging Face models (no API keys required).

Features

• PDF document reading and extraction
• Task planning and decomposition
• Summarization, keyword extraction, calculation, and question answering
• Critique and improvement of outputs
• Clean Streamlit UI for user interaction
• Download results as Markdown or PDF
• User feedback collection

Quick Start

1. Clone this repo and install requirements:

   pip install -r requirements.txt
   

2. Run the Streamlit app:

   streamlit run streamlit_app.py
   

3. Upload a PDF and enter any task or question (e.g., "Summarize the document", "What is the candidate's GPA?").

Example Tasks

• Summarize this resume and highlight key skills.
• What programming languages are listed?
• Calculate the mean of [85, 90, 78, 92].

Project Structure

• app.py - Main pipeline logic
• streamlit_app.py - Streamlit UI
• agents/ - Agent classes (Reader, Planner, Tools, Critic, Improver)
• tools/ - Tool implementations (summarizer, keywords, calculations, QA)
• utils/ - Utilities (chunking, postprocessing, export)

License

MIT