DOCUMIND-AI

Intelligent Offline Document Q&A Assistant

DOCUMIND-AI is an AI-powered offline document question-answering system that allows users to upload documents and ask natural language questions about their content. The system uses Retrieval-Augmented Generation (RAG) with local LLMs to ensure privacy, low latency, and zero dependency on cloud APIs.

This project focuses on practical AI engineering: document ingestion, semantic search, OCR, vector indexing, and end-to-end system integration.

---

🚀 Key Features

• 📄 Upload and process PDF, DOCX, and TXT documents
• 🔍 Semantic search using FAISS vector database
• 🧠 Context-aware Q&A using local LLMs (Ollama)
• 🖼️ OCR support for scanned PDFs using Tesseract
• 🔒 Fully offline & privacy-first architecture
• 📱 Mobile-friendly frontend built with React Native (Expo)

---

🧠 System Architecture

1. Document Ingestion Uploaded documents are parsed and split into chunks.

2. OCR Processing Scanned PDFs are processed using Tesseract OCR to extract text.

3. Embedding & Indexing Text chunks are converted into embeddings and stored in a FAISS index.

4. Query Processing User queries are embedded and matched against the FAISS index.

5. LLM Response Generation Relevant document context is passed to a local LLM via LangChain + Ollama.

---

🧩 Tech Stack

Backend

• Python
• Flask
• LangChain
• FAISS
• Ollama (local LLM runtime)
• Tesseract OCR

Frontend

• React Native
• Expo

---

📂 Project Structure

DOCUMIND-AI/
│
├── app.py                 # Flask backend entry point
├── ollama_llm.py          # Local LLM wrapper (Ollama + LangChain)
├── setup_models.py        # Script to setup required local models
├── utils.py               # Helper functions (OCR, embeddings, file handling)
│
├── document/              # Uploaded documents
├── index_store/           # FAISS vector indexes
│
├── frontend/              # React Native mobile application
│
├── requirements.txt       # Python dependencies
├── README.md              # Project documentation
└── LICENSE                # MIT License

---

📊 Design Decisions

• FAISS was chosen for fast, in-memory vector similarity search.
• Local LLMs (Ollama) ensure data privacy and offline usability.
• LangChain simplifies RAG pipeline orchestration.
• OCR integration enables handling of real-world scanned documents.

---

📈 Performance Notes

• Average query latency depends on model size and hardware
• Optimized for single-user, local inference
• Suitable for personal research, study, and document analysis

---

🎯 Use Cases

• Academic research paper analysis
• Resume and document review
• Legal or policy document exploration
• Personal knowledge base creation

---

🔮 Future Improvements

• Add unit and integration tests
• Improve chunking and retrieval accuracy
• Support multi-document conversation memory
• Add admin dashboard for document management
• Deploy backend as a containerized service

---

🧑‍💻 Author

Karan Shelar GitHub: https://github.com/Edge-Explorer

---

📜 License

This project is licensed under the MIT License.