🧘 Mandukya AI

Turn Videos Into Deep Understanding

AI-powered video study companion — your personal lecture assistant

---

🤯 Watched a 2-Hour Lecture and Remembered Nothing?

If you're serious about learning, you know the pain:

• ❌ Taking notes while watching is impossible
• ❌ Can't find that one concept from last week
• ❌ Re-watching hours of content to find a specific moment
• ❌ No way to search what was said in a video
• ❌ Your notes don't connect concepts together

What if AI could watch the videos for you, take perfect notes, and help you understand the big picture?

---

✨ What Mandukya AI Does

🎯 Automatic Transcription

Upload any video — lectures, YouTube, courses:

• Whisper AI speech-to-text (super accurate)
• Speaker identification
• Timestamped transcript
• Export as subtitles

Searchable video content.

📸 Smart Visual Capture

• Detects slide changes automatically
• Screenshots key diagrams and charts
• Captures important visual moments
• OCR extracts text from slides

See the highlights without watching.

🧠 Knowledge Graph

This is the magic:

• Extracts concepts from the content
• Connects related ideas together
• Shows you the "big picture"
• Find hidden relationships

Not just notes — actual understanding.

🎬 Interactive Presentations

• Auto-generated Reveal.js slides
• Transcript snippets with timestamps
• Jump to exact moments in video
• Review key points quickly

Navigate 2-hour lectures in 10 minutes.

🔍 Semantic Search

Ask questions like:

• "What did they say about consciousness?"
• "Find the part about the four states"
• "Show me all diagrams"
• "Summarize the main teaching"

AI understands the content, not just the words.

---

🔒 Privacy First

Your videos NEVER leave your computer:

• ✅ 100% local processing — No cloud, no subscriptions
• ✅ Your data stays yours — Everything on your machine
• ✅ Open source — Audit every line of code
• ✅ Offline capable — Works without internet

Perfect for sensitive or private content.

---

🚀 Get Started

Prerequisites

• Ollama installed and running
• Whisper.cpp compiled locally
• Python 3.10+

Quick Setup

# Clone
git clone https://github.com/Sensible-Analytics/video_analysis.git
cd video_analysis

# Setup
make setup

# Add your Whisper paths to .env

Usage Pipeline

Step 1: Download a Video

make download URL="https://www.youtube.com/watch?v=..."

Step 2: Process & Transcribe

make run

Step 3: Launch the Split-Helix UI

cd frontend
npm install
npm start

Step 4: Build Knowledge Graph

make index

Step 5: Search & Discover

make search QUERY="What are the main concepts?"

That's it. Deep understanding in minutes.

---

🎯 Perfect For

Students

"Upload lecture recordings. Get searchable notes. Study 3x faster."

Researchers

"Interview transcripts without manual transcription. Find quotes instantly."

Philosophy/Spiritual Seekers

"Deep study of philosophical texts — connect concepts across lectures."

Online Learners

"Coursera, YouTube tutorials, conference talks — make them all searchable."

---

🛠️ Architecture

Video Input → Whisper.cpp → Transcript
           → FFmpeg → Visual Frames
           ↓
     Cognee AI → Knowledge Graph
           ↓
     Ollama LLM → Insights & Diagrams
           ↓
     Split-Helix UI → Interactive Experience

Privacy-first. Local processing. Deep understanding.

---

💡 What Makes This Different

Other video tools:

• Upload to their servers ❌
• Monthly subscription ❌
• Basic transcripts only ❌
• No concept connections ❌

Mandukya AI:

• Runs on YOUR computer ✅
• Free forever ✅
• Knowledge graphs + transcripts ✅
• Connects concepts automatically ✅

---

🎓 Originally Built For

This tool was created for deep study of the Mandukya Upanishad — exploring the four states of consciousness (waking, dreaming, deep sleep, pure consciousness).

But it works beautifully for ANY educational content:

• University lectures
• YouTube tutorials
• Conference talks
• Online courses
• Interview recordings

---

🛠️ Tech Stack

• Whisper.cpp — Local speech recognition
• Ollama — Local LLM for insights
• Cognee — Knowledge graph generation
• FFmpeg — Video processing
• Python — Backend processing
• React — Split-Helix UI

---

🤝 Built By

Sensible Analytics — AI that respects your privacy

Want custom AI learning tools? Let's talk.

---

Start understanding deeper.

🚀 Get Started · ⭐ Star on GitHub

---

🚀 Getting Started (Technical Setup)

1. Prerequisites

• Ollama installed and running.
• Whisper.cpp compiled locally.
• Python 3.10+

2. Setup

make setup

This will install dependencies and create your .env file. Edit .env to point to your Whisper binary and model.

3. Usage Pipeline

Step A: Download Lessons

make download URL="https://www.youtube.com/playlist?list=..."
make run  # Start the background brain
cd frontend && npm install && npm start # Launch the Split-Helix UI

and generates Reveal.js slides in the slides/ directory.

Step C: Build Knowledge Graph

make index

Uses Cognee to extract entities and relationships across all lessons, building your local RDBMS, Vector, and Graph databases.

Step D: Semantic Discovery

make search QUERY="The four states of consciousness"

🏗️ Architecture

The system follows a modular "Knowledge Extraction" architecture:

1. Perception: Whisper.cpp (Audio -> Text) & FFmpeg (Video -> Frames).
2. Memory: Cognee (Text -> RDBMS/Vector/Graph).
3. Reasoning: Ollama (Context + Chunk -> Insights/Diagrams).
4. Presentation: Reveal.js (Data -> UI).

---

Created for the study of Mandukya Upanishad.