Amharic-English Translation AI App

I want to create an AI-driven application that translates seamlessly between Amharic and English. The core of the product is high-quality language translation, but it must also include a full voice-recognition layer:

• Speech-to-text conversion so users can dictate Amharic or English and see accurate transcripts.
• Voice-command execution to let them control key in-app actions hands-free.
• Language identification that automatically detects whether the incoming audio or text is Amharic or English before processing.

The finished app should run smoothly on Android phones and as a responsive web application; a shared code-base or robust API that feeds both fronts is ideal. I’m open to your preferred stack—TensorFlow, PyTorch, Whisper, or any other modern NLP/ASR tools—as long as model performance stays fast and the architecture can be retrained with additional corpora later.

Deliverables
1. Source code with clear build instructions.
2. A runnable Android APK and a hosted web demo.
3. Documentation covering model training steps, data requirements, and how to modify voice commands.
4. Brief performance report that lists translation BLEU scores and word-error rates for speech-to-text in both languages.

Please share any relevant Amharic NLP work you’ve done, outline the libraries you plan to use, and propose a timeline for prototype, testing, and final hand-off.