Gestura is
fully open source.
Every line of code — from the LSTM training pipeline to the React frontend — is publicly available on GitHub. Fork it, learn from it, build on it.
Tech stack
Everything built from scratch.
PyTorch + LSTM
Custom LSTM neural network trained on 87,000 ASL images. Runs on Apple Silicon via MPS acceleration.
MediaPipe
Extracts 21 hand landmarks per frame. We use normalized coordinates — not raw pixels — for faster, lighter inference.
FastAPI + Uvicorn
Python REST API serving predictions. Processes each webcam frame in under 200ms with full CORS support.
React + Next.js
Modern frontend with live webcam feed, hand landmark overlay, confidence graph, and real-time sentence builder.
Docker + GitHub Actions
Fully containerized with a CI/CD pipeline. Push to main and it deploys automatically to AWS.
AWS EC2 + Vercel
Backend runs on AWS EC2 inside Docker. Frontend deployed on Vercel with a custom domain.
Contribute
Want to help build Gestura?
Report a bug
Found something broken? Open an issue on GitHub and we'll fix it fast.
Suggest a feature
Have an idea for improving Gestura? We'd love to hear it. Open a discussion.
Improve the model
Have a better dataset or training approach? Submit a PR and let's improve accuracy together.
Spread the word
Star the repo, share it on LinkedIn, or write about it. Every bit helps.