🧠 DocuMind AI

Intelligent document analysis and Q&A — powered by Google Gemini

---

🔗 Live Demo

👉 https://docu-mind-vansh.streamlit.app/

---

📌 What is DocuMind AI?

DocuMind AI is a production-grade Agentic RAG (Retrieval-Augmented Generation) system that lets you chat with PDFs, websites, and YouTube videos using AI.

It doesn't just answer questions — it:

• Thinks before responding
• Reviews its own answers
• Adapts based on confidence
• Compares documents
• Extracts timelines
• Generates flashcards

---

✨ Features

💬 Chat with Any Content

Source	Input Method
📄 PDF	Upload one or multiple PDFs
🌐 Website	Paste any URL
▶ YouTube	Paste video link

---

🤖 3 Agentic AI Layers

Agent	Role
Clarification Agent	Detects vague queries (no API cost)
Self-Reflection Agent	Improves generated answers
Confidence Adaptation Agent	Adds uncertainty warnings

---

🧬 Document DNA

Auto-generated document fingerprint including:

• Domain
• Tone
• Complexity
• Sentiment
• Key themes
• Entities
• Summary
• Unique insights

---

🔀 Doc vs Doc Comparison

• Dual FAISS retrieval

• Color-coded sources

  • 🟢 Doc A
  • 🔵 Doc B

---

🕐 Timeline Extractor

• Extracts dates, deadlines, milestones
• Visual structured timeline

---

🃏 Flashcard Generator

• Auto Q&A cards
• Difficulty filters (Easy / Medium / Hard)

---

🛠 Smart Tools

• Auto Summary
• Quiz Generator
• Email Drafter
• Contradiction Finder
• Action Extractor

---

📊 Session Analytics

• Confidence trends
• Keyword frequency
• Low-confidence detection

---

🌐 Multilingual Support

• Auto language detection
• Responses in same language

---

🏗 Project Structure

documind/
├── app.py
├── utils/
│   ├── styles.py
│   ├── pdf_processor.py
│   ├── ai_helpers.py
│   └── agents.py
└── components/
    ├── tab_chat.py
    ├── tab_dna.py
    ├── tab_tools.py
    ├── tab_analytics.py
    ├── tab_compare.py
    ├── tab_timeline.py
    └── tab_flashcards.py

---

🔧 RAG Pipeline

Input (PDF / URL / YouTube)
→ Text Extraction
→ Chunking (600 chars, 80 overlap)
→ Embedding (Gemini Embedding 001)
→ FAISS Index

Query
→ Clarification Agent
→ Similarity Search (Top-K)
→ Confidence Scoring
→ Answer Generation (Gemini 2.5 Flash)
→ Self-Reflection
→ Confidence Adaptation
→ Final Response + Sources

---

---

📦 Tech Stack

Technology	Purpose
Python 3.12	Core
Streamlit	UI
LangChain	AI orchestration
Gemini 2.5 Flash	LLM
Gemini Embedding	Vector generation
FAISS	Vector search
PyPDF	PDF parsing
BeautifulSoup	Web scraping
youtube-transcript-api	Video transcripts
Web Speech API	Voice input

---

👨‍💻 Author

Vansh Mahajan

• 📧 vansh150705@gmail.com
• 💼 https://www.linkedin.com/in/vansh-mahajan-napv/
• 🐙 https://github.com/Vansh150705

---

Built with 🧠 by Vansh Mahajan