Show HN: Tuby.dev – Indexing Rails videos via Vision AI code analysis(tuby.dev)
tuby.dev
Show HN: Tuby.dev – Indexing Rails videos via Vision AI code analysis
https://tuby.dev/
2 コメント
Curious about the OCR + Vision pipeline: do you run frame sampling, or analyze every frame? How do you deal with false positives when extracting gem names or code snippets?
I built Tuby.dev to solve this. It's a curated aggregator for Ruby/Rails videos.
The Engineering:
Instead of relying on YouTube's metadata or captions (which are often incomplete), I built a pipeline that:
1. Downloads the video locally (to avoid datacenter blocking). 2. Sends it to Gemini 1.5 Flash (Vision API). 3. Performs OCR on the code shown on screen.
This allows the system to index Gems, patterns, and versions that the author used in the code but forgot to mention in the description or audio.
The Stack: It's running on Rails 8 with Inertia.js (React). Ruby version is 3.4.7 with Prism parser enabled.
I'd love to hear your thoughts on the classification approach or the stack.