AI Video Code Extractor

The problem

Tutorials are great for learning, but transcribing code from a video into your editor is friction. I wanted a single button that turned a 20-minute video into a deduplicated list of code blocks.

The approach

• OpenCV scans frames and detects scene changes where the IDE content shifts.
• Tesseract OCR with a monospace-tuned PSM mode pulls text from those frames.
• Gemini cleans OCR noise, merges adjacent frames into single snippets, and identifies the language for syntax highlighting.
• Streamlit UI lets users upload a video and see snippets appear progressively.

Outcome

• Accurate on screencasts; works even on stylized fonts at 1080p.
• Demonstrated the value of pairing a cheap OCR pass with an LLM cleanup step instead of relying on a multimodal model end-to-end.