Back to catalogue
Project
Hospital AI Translation
Translation prototype serving Yoruba, Nupe, and English through a Flask API and a Flutter client.
How it’s builtHow it’s built
Hospital AI Translation · Doctor-patient translation prototype for Yoruba, Nupe and English
I built a translation prototype for doctor-patient conversations: a Flask API serving three fine-tuned MarianMT models that I published on Hugging Face (Yoruba to English, Nupe to English, Nupe to Yoruba), and a Flutter client that posts text to the API. The repo includes a 144-sentence trilingual clinical phrase dataset.
- MarianMT translation models served
- 3
- parallel Yoruba/Nupe/English sentences
- 144
- Flask API endpoint (POST /translate)
- 1
- lines in the Flutter client
- 237
Technology and what it does here
- Hugging Face Transformers (MarianMT)
- Loads three fine-tuned MarianMTModel/MarianTokenizer checkpoints from the gabri3l namespace
- PyTorch
- Runtime for model.generate during translation
- Flask + flask-cors
- POST /translate routes src_lang-tgt_lang pairs to the right model
- Flutter / Dart
- Mobile client with backend URL screen, language dropdowns and translator screen
- http (Dart package)
- Posts JSON to the Flask /translate endpoint
- Windows batch scripts
- Run.bat checks Python 3.11, installs requirements and starts Flask on port 8090
Implementation details
- The API dispatches on a 'src-tgt' key (yo-en, nu-en, nu-yo) to one of three models loaded once at startup.
- The dataset holds clinical phrases such as 'Good morning doctor, I am not feeling well.' aligned across all three languages.
- A prebuilt Android build is committed as Hospital-AI.apk in the backend repo.
Technologies
- Translation
- Flask
- Flutter
