Spaces:
Running on Zero
Running on Zero
File size: 3,377 Bytes
cfeec66 30d966c cfeec66 30d966c cfeec66 30d966c cfeec66 c0da4af d262a06 30d966c c0da4af 30d966c c0da4af d262a06 30d966c c0da4af 131bc36 30d966c c0da4af 30d966c c0da4af 30d966c c0da4af c8021f8 d262a06 c0da4af c8021f8 c0da4af d262a06 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 | ---
title: G-MASS - Ghana Medical AI Safety Screen
emoji: 🏥
colorFrom: blue
colorTo: yellow
sdk: gradio
sdk_version: 5.23.1
app_file: app.py
pinned: true
license: apache-2.0
tags:
- medical-ai
- ai-safety
- ghana
- twi
- african-languages
- llm-evaluation
- health
- nlp
short_description: Medical AI safety eval for Ghanaian languages
---
# G-MASS: Ghana Medical AI Safety Screen
**MediSafe-GH · Biomedical Technologies Lab**
G-MASS evaluates whether AI health assistants respond safely to clinical queries across **English**, **Ghanaian English**, and **Twi**.
---
## 🚀 How to Use the Interface
1. **Single Probe Evaluation**: Enter a medical question in English, Ghanaian English, or Twi, choose your target model, and evaluate for immediate safety verdicts, language detection, and clinical referral adequacy.
2. **Batch Evaluation**: Upload `.jsonl`, `.csv`, `.ndjson`, or `.json` datasets to run evaluations across entire probe sets and download scored CSV reports.
3. **Benchmark Results**: Inspect empirical Clinical Safety Rates (CSR) and Cross-Lingual Safety Degradation Scores (SDS).
4. **Settings & Compute Tiers**: Enter custom session API keys, adjust SDS deployment thresholds, or toggle between judge compute tiers.
5. **Community & Issue Tracker**: Submit clinical safety hazard reports, flag false positives or Twi dialect nuances, and open direct GitHub Issues or Pull Requests.
6. **Contact & Support**: Reach out to the research team directly at `biomedicaltechnologieslab@gmail.com`.
---
## ⚙️ Compute Tiers
G-MASS supports adaptive compute scaling:
- **Tier 1 (Nano)**: CPU-only FastText + lightweight rule heuristics (~0.3s/probe).
- **Tier 2 (Standard - Default)**: LlamaGuard3-1B + AfroLM ensemble (~1–2s/probe).
- **Tier 3 (Heavy)**: 16GB+ VRAM GPU, full LlamaGuard3-8B / Gemma3-7B research ensemble.
- **Tier 4 (API-only)**: Zero local compute, hosted cloud policy judge.
---
## 🔑 Required & Optional API Keys
- `GEMINI_API_KEY`: Gemini 2.5 Flash evaluation and hosted policy judge.
- `OPENAI_API_KEY`: GPT-4o / GPT-4o mini evaluations.
- `HF_TOKEN`: Hugging Face router / open-weight models (Phi-3, BioMistral).
- `KHAYA_API_KEY`: Real-time Khaya / GhanaNLP translation.
> **Note**: Custom keys can be entered directly in the **Settings** tab for individual sessions without exposing secrets.
---
## ⚠️ Disclaimer
G-MASS is an experimental research prototype developed by the Biomedical Technologies Lab for AI benchmarking and evaluation purposes only. It does not provide clinical diagnosis, medical advice, or formal medical device certification.
---
## 📬 Contact & Support
- **Email**: [biomedicaltechnologieslab@gmail.com](mailto:biomedicaltechnologieslab@gmail.com)
- **GitHub**: [Armstrong66/medisafe-gh](https://github.com/Armstrong66/medisafe-gh)
- **Space**: [BioinstLab/gmass-demo](https://huggingface.co/spaces/BioinstLab/gmass-demo)
- **Institution**: Biomedical Technologies Lab
---
## 📄 Citation
```bibtex
@software{medisafe_gh_2026,
author = {Koduah, Joseph Derrick Anane Nti and Asare, Michael Asiedu and Owusu, Emmanuel and Yeboah, Benjamin Appiah},
title = {G-MASS: Ghana Medical AI Safety Screen},
year = {2026},
publisher = {Hugging Face},
institution = {Biomedical Technologies Lab},
url = {https://huggingface.co/spaces/BioinstLab/gmass-demo}
}
```
|