I need an end-to-end, browser-based MVP that can translate spoken or written English into real-time American Sign Language and vice-versa. The core experience will run on desktop first, so please optimize the UI and performance for larger screens and full keyboard/mouse interaction. Scope of work • Build a production-ready ASL recognition engine that listens to webcam input or typed text and converts it into sign language. • Create a signing engine that drives a life-like 3-D avatar rendered in the browser (WebGL / Three.js or comparable) to display the translation fluidly. • Deliver a polished UI/UX, from onboarding through live conversation, with React (or similar) on the front end. • Implement a fully customizable avatar system. Users should be able to adjust facial traits, skin tone, hair, and clothing so the signer can resemble family, friends, or themselves. • Lay the groundwork for adding more sign languages. No specific language beyond ASL is locked in yet, so structure the architecture for easy model swapping. • Integrate facial-expression and emotion detection so the avatar conveys nuance, not just hand shapes. • Enable two-way live communication (Deaf ↔ hearing) using WebRTC or another low-latency protocol. • Provide a training pipeline (Python / TensorFlow) that lets me improve accuracy and reduce latency over time with user-approved datasets. Acceptance criteria 1. MVP loads in the latest Chrome and Edge on desktop, achieving <300 ms end-to-end latency for a simple sentence. 2. Avatar signs a supplied English phrase with correct hand shapes, facial grammar, and timing. 3. User can customize and save an avatar profile in under two minutes without external tools. 4. Architecture documents show how additional sign-language models can be plugged in with minimal refactor. 5. Codebase is clean, commented, and delivered via a private Git repository with setup instructions. I’m happy to discuss preferred stacks or libraries if you have strong opinions that meet these goals.