The latest articles on DEV Community by Anton Zaharenko (@anton_zaharenko). https://dev.to/anton_zaharenko https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4008716%2Fde67ba1b-34e0-4c93-be05-b908a14f2032.jpg https://dev.to/anton_zaharenko en Anton Zaharenko Tue, 30 Jun 2026 08:25:51 +0000 https://dev.to/anton_zaharenko/i-built-a-chrome-extension-nn0 https://dev.to/anton_zaharenko/i-built-a-chrome-extension-nn0 <p>I built a Chrome extension that extracts code from tutorial <br> videos using Claude's Vision API. Wrote up what I learned:</p> <p>Key findings:</p> <ol> <li><p>Vision models handle code extraction surprisingly well better than OCR for anything syntax-sensitive</p></li> <li><p>Prompt engineering matters more than model choice. <br> "Extract the code" vs "identify all code blocks, preserve <br> indentation, differentiate from terminal output" huge <br> difference in quality</p></li> <li><p>Latency: ~2-3 seconds for a 1080p frame. Acceptable for <br> this use case, too slow for real-time</p></li> <li><p>Edge case: slides with code + screenshots of code — <br> model correctly identifies both as "code" even with <br> different visual styling</p></li> </ol> <p>Full writeup with prompt examples: [link to dev.to article]</p> <p>The extension itself is free if anyone wants to try the <br> output quality: <a href="https://codera.click/" rel="noopener noreferrer">codera.click</a></p> ai claude showdev webdev