Graduation Project & Recognition
My Contribution
- Built the FastAPI and PostgreSQL architecture, data model, and backend services that connect Arabic input to ordered sign-playback data.
- Integrated WebSocket audio, Amazon Transcribe, and local Whisper for online and offline speech recognition.
- Implemented a ten-stage Arabic processing pipeline combining lexical matching, morphological rules, and constrained large language model fallback. Unknown inputs remain explicit when the available vocabulary cannot resolve them.
- Built 65 unit tests covering Arabic processing, semantic guardrails, session logic, playback, and video-export helpers.




















