🧠 DocuMind AI: Simple Document Chat

Chat with your PDFs using AI. This project uses Gemini 1.5 to answer questions based on your uploaded documents with high accuracy and cited sources.

---

✨ How It Works (The Simple Version)

1. 📤 Upload: Your PDF is chopped into small, smart pieces to maintain context.
2. 💾 Store: Those pieces are converted into "embeddings" (numerical representations) and saved in a Vector Database.
3. 💬 Ask: When you ask a question, the AI finds the most relevant pieces and writes a clear answer based only on them.

---

🚀 Key Features

• Smart Document Processing: Layout-aware PDF handling with semantic chunking for zero-context loss.
• Vector Database (ChromaDB): Optimized storage for sub-second document retrieval and semantic ranking.
• Strict Context Grounding: Gemini 1.5 Flash is forced to answer only from your uploaded context—eliminating AI hallucinations.
• Modern Dark Mode UI: A sleek, minimal dashboard built with Next.js 14 and Glassmorphism Styling.
• Interactive Citations: Provides the exact filename and page number for every AI-generated claim.

---

🏗️ Architecture

graph TD
    A[Browser: Next.js Dashboard] -- "Query (JSON)" --> B[FastAPI: Intelligence Bridge]
    B -- "1. Cleanse Input" --> C{Neural Core}
    C -- "2. Vector Search" --> D[Vector Store: ChromaDB]
    D -- "3. Relevant Segments" --> C
    C -- "4. Reason with Context" --> E[AI: Gemini 1.5 Flash]
    E -- "5. Formulate Citation" --> B
    B -- "6. Verified Answer" --> A

Loading

---

📂 Project Structure

documind-ai/
├── backend/
│   ├── app/
│   │   ├── services/
│   │   │   ├── ingestion.py      # PDF parsing logic
│   │   │   └── rag_service.py    # AI reasoning & retrieval
│   │   ├── main.py               # API endpoints
│   │   └── models.py             # Data schemas
│   ├── Dockerfile                # Backend container config
│   └── requirements.txt          # Python dependencies
│
├── frontend/
│   ├── app/                      # Next.js 14 Dashboard
│   ├── globals.css               # Premium styles
│   ├── package.json              # NPM manifest
│   └── tailwind.config.ts        # UI theme config
│
├── docker-compose.yml            # Multi-service setup (Local)
├── .env.example                  # Environmental template
└── README.md                     # Project documentation

---

🛠️ Setup Guide

1. Prerequisites

• Node.js 20+ and Python 3.11+
• Google AI API Key: Get it at Google AI Studio

2. 🔑 Environment Configuration (Railway Backend)

When deploying your Backend to Railway, ensure you add the following Variables:

• GOOGLE_API_KEY: Your key from Google AI Studio.
• CHROMA_DB_PATH: Set to /app/vector_store for production persistence.
• PORT: 9000

🚀 Persistence: In your Railway Dashboard, go to Settings -> Volumes and mount a volume at /app/vector_store. This ensures your indexed documents are not lost after an app restart.

---

🆘 Troubleshooting

• PDF Parsing Errors: If a PDF is a scanned image, ensure you are using a text-based PDF or use an OCR library like PyMuPDF.
• Port Conflicts: If Port 3000 is taken, the frontend will fail to start. Change the port in docker-compose.yml if needed.
• Backend Connection: If the AI doesn't respond, ensure the NEXT_PUBLIC_API_URL in your .env matches your running backend port.

---

Stack: FastAPI · LangChain v0.2 · Gemini 1.5 · Next.js 14 · ChromaDB · Tailwind CSS · Docker

Developed as a clean, accessible solution for semantic document research.