Spaces:
Sleeping
title: VoxDoc Voice Symptom Triage Assistant
colorFrom: blue
colorTo: green
sdk: docker
app_port: 7860
pinned: false
license: apache-2.0
short_description: Voice intake and clinical documentation demo
thumbnail: >-
https://cdn-uploads.huggingface.co/production/uploads/696a57ff06f0d3d8e09b7433/whTfF2UGXkXM5zhjWvt46.png
VoxDoc β HF Space Deployment
Demo deployment of voice-symptom-triage-assistant on Hugging Face Spaces.
Demo only. Hugging Face Spaces are not HIPAA-compliant. Do not submit real protected health information (PHI). Use synthetic transcripts only.
How this Space is configured
This Space runs the FastAPI backend on the free CPU tier. To stay within
the 2-vCPU / 16-GB RAM budget, MedGemma generation is delegated to HF
Inference Providers rather than loaded in-process. ASR (whisper-small)
and biomedical NER still run locally on CPU.
| Setting | Value | Why |
|---|---|---|
MEDGEMMA_PROVIDER |
hf-inference |
Avoid loading the 4b model on CPU |
DEVICE / ENABLE_GPU |
cpu / false |
Free tier has no GPU |
ENABLE_IMAGE_ANALYSIS |
false |
Vision routing not configured here |
app_port |
7860 |
HF Spaces default |
Required Space secrets
Add these in Settings β Variables and secrets β New secret:
HF_TOKENβ read-token from huggingface.co/settings/tokens with access to MedGemmaENCRYPTION_MASTER_KEYβ any strong random string (openssl rand -hex 32)JWT_SECRET_KEYβ same; used even with auth disabled
Optional: persist sessions across restarts
The default SQLite store lives on the ephemeral Space filesystem and is lost on restart. To keep sessions, either:
- Enable persistent storage (Settings β paid feature) and point
the SQLite file at
/data/voxdoc.db, or - Set
DATABASE_URLto a free hosted Postgres (Neon / Supabase) using thepostgresql+asyncpg://...form.
Frontend
The React frontend (frontend/) is not built into this Space β deploy it
separately to Vercel/Netlify and point its API base URL at this Space's URL.
The legacy server-rendered UI under app/static/ is still reachable for
quick testing.
Local equivalent
docker build -f deploy/huggingface/Dockerfile -t voxdoc-hf .
docker run -p 7860:7860 \
-e HF_TOKEN=hf_xxx \
-e ENCRYPTION_MASTER_KEY=$(openssl rand -hex 32) \
-e JWT_SECRET_KEY=$(openssl rand -hex 32) \
voxdoc-hf
Open http://localhost:7860.