AI Research Assistant 🤖

An advanced, LLM-powered agent designed to deconstruct and simplify complex academic research papers. This tool takes a PDF as input, generates a clean, structured booklet with summaries and citations, and provides an interactive chat interface for Q&A.

This project was developed for the Intra IIT Tech Meet at IIT Bhilai.

---

Features ✨

• Intelligent Document Parsing: Uses unstructured to parse PDFs, extracting text, tables, and images.
• Automated Booklet Generation: Creates a professional, multi-page PDF booklet containing:
  • A concise, AI-generated summary of the paper.
  • A list of generated citations for key concepts.
  • Extracted tables and image captions.
• Stateful Agentic Workflow: Built with LangGraph, using a supervisor architecture to robustly manage the processing pipeline, chat interactions, and error handling.
• Interactive RAG Chat: After processing, a chat interface is enabled, allowing users to ask specific questions about the paper's content.
• Modular & Asynchronous Backend: Built with FastAPI to handle requests efficiently, offloading heavy processing to background tasks.

---

Architecture Overview

The application uses a state-machine architecture managed by a LangGraph "supervisor." This ensures a clear separation between the document processing pipeline and the interactive chat loop, providing a smooth and predictable user experience.

graph TD
subgraph "Phase 1: Document Processing Pipeline"
    A[User uploads PDF via UI/API] --> B{FastAPI Endpoint: /process_document};
    B -- Triggers --> C{LangGraph Workflow};
    C --> D(1. Parse Document);
    D --> E(2. Index Content for RAG);
    E --> F(3. Summarize Paper);
    F --> G(4. Generate Citations);
    G --> H(5. Compile PDF Booklet);
    H --> I((State Updated: Chat Ready ✅));
    end

subgraph "Phase 2: Interactive Chat Loop"
    J[User asks question via UI/API] --> K{FastAPI Endpoint: /chat};
    K -- Uses RAG Index from E --> L[RAG Chain Retrieves Context];
    L --> M[LLM Generates Answer];
    M --> N((Response Sent to User));
end

---

Tech Stack 🛠️

• Backend: FastAPI
• AI Orchestration: LangGraph
• LLM Interaction: LangChain, Google Gemini
• Document Parsing: unstructured
• Vector Database: ChromaDB
• Embeddings: HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
• PDF Compilation: WIP

---

Setup and Installation ⚙️

Follow these steps to set up and run the project locally.

1. Clone the Repository

git clone <your-repo-url>
cd <your-repo-name>

2. Create a Virtual Environment

# For macOS / Linux
python3 -m venv venv
source venv/bin/activate

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

3. Install System Dependencies

This project requires two key system-level dependencies:

• Poppler (for unstructured):
  • macOS: brew install poppler
  • Debian/Ubuntu: sudo apt-get install poppler-utils
• LaTeX (for PDF compilation):
  • macOS: MacTeX
  • Windows: MiKTeX
  • Debian/Ubuntu: sudo apt-get install texlive-latex-base texlive-fonts-recommended texlive-latex-extra

4. Install Python Packages

Create a requirements.txt file with the content below and install it.

Got to requirements.txt file and copy paste

pip install -r requirements.txt

5. Set Up Environment Variables

Create a file named .env in the project root directory and add your Google API key.

# .env
GOOGLE_API_KEY="your_google_api_key_here"

---

How to Use 🚀

1. Start the Backend Server: Run the following command from the project's root directory:

   uvicorn app.main:app --reload

   The API will be available at http://127.0.0.1:8000.

2. Access the Interactive API Docs: Open your browser and navigate to http://127.0.0.1:8000/docs.

3. Process a Document:

   • Expand the /process_document endpoint.
   • Click "Try it out" and upload a PDF file.
   • Execute the request. You will receive a job_id.

4. Check Status & Get Booklet:

   • Expand the /status/{job_id} endpoint.
   • Enter the job_id from the previous step and execute.
   • Once the current_mode shows "chatting", a booklet_url will be available to download the generated PDF.

5. Chat with the Document:

   • Expand the /chat endpoint.
   • Enter the same job_id and your question in the request body.
   • Execute to receive an AI-generated answer based on the document's content.

---

Project Structure

/project_root
├── app/
│   ├── agents/
│   │   ├── summarizer.py
│   │   └── citation_generator.py
│   ├── graphs/
│   │   ├── supervisor.py
│   │   ├── processing_nodes.py
│   │   ├── chat_nodes.py
│   │   └── error_nodes.py
│   ├── utils/
│   │   ├── parser.py
│   │   └── pdf_compiler.py
│   ├── schemas.py
│   └── main.py
├── outputs/        # Generated booklets and images will appear here
├── templates/
│   └── booklet_template_v2.tex
├── .env
└── requirements.txt