LLMFlow: Summarization of Documents

LLMFlow integrates GPT with LangChain, DeepSeek and LLama3.2 with Ollama to generate concise summaries of large-scale documents (research documents), reducing reading time and effort for academics. The project enhances accessibility and efficiency in reviewing literature.

This project allows users to upload a research paper or a document (word, txt, and PDF), generate a summary, and ask questions about its content using various Large Language Models (LLMs) such as Ollama, GPT, and DeepSeek.

Features

• Upload document – Supports .txt``.docx``.pdf format.
• Generate Summaries – Extracts key sections and provides a concise summary.
• Chat with the Paper – Ask questions and get AI-generated responses based on the document.
• Multiple LLMs Supported – Choose from Llama, GPT, or DeepSeek for text processing.
• Context-Aware Responses – Retrieves the most relevant sections from the document before answering.

App Screenshot

---

Technologies Used

Frontend (React)

• React.js – For building the interactive chat interface.
• Tailwind CSS – For styling and responsive design.
• JavaScript (ES6+) – Used for frontend logic.
• Fetch API – For making requests to the backend.
• React Hooks (useState, useEffect, useRef) – For managing state and interactions.

Backend (FastAPI)

• FastAPI – A modern Python web framework for handling API requests.
• Uvicorn – ASGI server to run the FastAPI app.
• Pydantic – Data validation for request bodies.
• CORS Middleware – Enables cross-origin requests.

Large Language Models (LLMs)

• Ollama – Runs local LLMs like Llama3.2 for chat responses.
• GPT (OpenAI GPT-4o) – (Optional) Used via LangChain for advanced processing.
• DeepSeek – (Optional) Another LLM used for extraction.

File Processing

• PDFMiner – Extracts text from uploaded PDFs.
• FuzzyWuzzy – For text similarity matching to find relevant document sections.
• Regular Expressions (Regex) – To detect and structure research paper sections.

Additional Tools & Libraries

• shutil & os – For file handling.
• Logging – For error tracking and debugging.

---

Installation Guide

Clone the Repository

git clone https://github.com/sabdulrahman/LLMFlow.git
cd LLMFlow

Set Up the Backend (FastAPI)

Create and Activate a Virtual Environment

python -m venv venv
source venv/bin/activate  # On Windows use `venv\Scripts\activate`

Install Dependencies

pip install -r requirements.txt

Run the FastAPI Server

On windows open the run.bat file.

uvicorn main:app --reload

The backend will be running at: http://localhost:8000

---

Set Up the Frontend (React)

cd frontend
npm install
npm start

The frontend will be available at: http://localhost:3000

---

Usage

1. Open the web interface in your browser at http://localhost:3000.
2. Upload a PDF research paper.
3. The system will generate a summary of the document.
4. Type your question related to the document in the chat.
5. The system retrieves relevant sections and generates an AI-powered response.

---

API Endpoints

Method	Endpoint	Description
POST	/upload-file	Upload a research paper (PDF) and extract sections.
POST	/process-message	Process user queries with the selected LLM.
GET	/	Check if the backend is running.

---

Environment Variables

To use GPT-based processing, create a .env file in the backend directory and add:

OPENAI_API_KEY=your_openai_api_key

---

Future Improvements

• Enhanced Document Processing – Better PDF parsing and section extraction.
• Multi-Document Support – Upload and interact with multiple documents.
• Advanced Query Matching – Improve accuracy in retrieving document sections.

---

Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request.

License

MIT License. See LICENSE for more details.