📄 Doc-Fusion

Doc-Fusion is a robust, intelligent application that transforms collections of academic or technical PDFs into semantically searchable and structured outputs—including LaTeX-based review papers—with minimal effort. Powered by LlamaParse, Google Gemini, and Milvus, Doc-Fusion extracts, summarizes, and indexes documents into a vector database, enabling natural language querying and automated review generation.

---

🚀 Features

• 🔍 AI-powered PDF extraction and chunking using LlamaParse
• 🤖 Summarization via Google Gemini
• 🧠 Keyword-driven semantic similarity search
• 📚 Embedding & vector search with Milvus
• 🧾 Automated LaTeX review paper generation
• 📄 Compile LaTeX into polished, formatted PDF outputs
• 🌐 Streamlit web app interface
• 🐳 Full Docker container support
• 📂 Works with multiple PDFs
• 🧱 Modular architecture for easy scaling and extension

---

🧠 How It Works

1. Multi-PDF Ingestion
   Upload multiple PDF documents as raw inputs.

2. Keyword-Based Similarity Search
   A user-defined keyword query drives a semantic similarity search, extracting the most relevant content from the dataset.

3. Dynamic LaTeX Generation
   Retrieved content is automatically converted into structured LaTeX snippets—including sections, figures, citations.

4. Review Paper PDF Output
   Compiled LaTeX is rendered into a professionally formatted review paper with customized styling.

---

⚙️ Installation Steps

1. Clone the repository

git clone https://github.com/vidhi-github/doc-fusion.git
cd doc-fusion

2. Generate API Keys

• LlamaParse (LlamaIndex Cloud): https://cloud.llamaindex.ai
• Google Gemini: https://makersuite.google.com/app

3. Create a .env file in the root directory and paste:

LLAMA_CLOUD_API_KEY=your_llama_key
GEMINI_API_KEY=your_gemini_key

4. Create and activate a virtual environment

python -m venv myenv
myenv\Scripts\activate       # On Windows

5. Install dependencies

pip install -r requirements.txt

6. Parse and dump PDFs

python automation.py dump data/cnn1.pdf data/cnn2.pdf output

7. (Optional) Run with Docker

docker compose up -d

8. Search using a query

python automation.py search "Convolutional Neural Network"

9. Launch Streamlit app

streamlit run app.py

---

📁 Directory Structure

Doc-Fusion/
│
├── automation.py              # Main automation script for dumping/searching PDFs
├── app.py                     # Streamlit app interface
├── requirements.txt           # Python dependencies
├── .env                       # API keys (create manually)
│
├── data/                      # Input PDFs
├── output/                    # JSON summaries
├── myenv/                     # Virtual environment
├── static/                    # Optional LaTeX assets
│
├── Dockerfile                 # Docker configuration
├── docker-compose.yml         # Docker service runner
│
├── parser.py                  # LlamaParse-based PDF parser
├── retrieval.py               # Vector DB handling (Milvus)
├── ToLatex.py                 # LaTeX generator module
├── usegemini.py               # Gemini summarizer
├── llm_prompt.py              # Prompt templates for LLMs
└── README.md                  # You're reading it!

---

🌍 Applications

• 🧠 Academic Research: Streamlined review paper generation and literature analysis.
• 📚 Document Management: Ideal for law firms, universities, or corporate knowledge bases.
• 🔍 AI-Powered PDF Search Engines: Semantically index and search documents using natural queries.
• 🏥 Healthcare & Legal: Summarize case files, reports, or records with AI.
• 💼 Enterprise Knowledge Extraction: Convert legacy documentation into structured insights.

---

🛠️ Tech Stack

• Python 3.12
• LlamaParse (LlamaIndex Cloud)
• Google Gemini
• Milvus (Vector DB)
• PyMuPDF (for layout-preserving parsing)
• Streamlit
• Docker

---

🚀 Future Scope

• 🔧 Drag-and-drop GUI for PDF uploads
• 🤖 Voice-enabled chatbot for reading summaries aloud
• 🌐 Multilingual support for translation and search
• 📄 OCR support for scanned PDFs
• ☁️ Cloud upload (AWS S3, Firebase, etc.)
• 📊 Analytics dashboard for document insights
• 🧩 Plug-in support for other LLMs (GPT-4, Claude, etc.)
• 📱 Cross-platform mobile app
• 🧠 Fine-tuned models for domain-specific outputs
• 🧪 Auto-evaluation of summary quality and key insights

---

🙌 Contributing

Pull requests are welcome!
To contribute:

# Fork the repo
git checkout -b feature/your-feature
# Make your changes
# Submit a pull request to the main branch

---

📬 Contact

Built with ❤️ by Vidhi Jindal
Feel free to reach out for collaborations or feedback! Thank You for going throug my project.

---