Omdena — VisionVitals
Computer vision health monitoring app combining Django, FastAPI, MediaPipe, and PyTorch in a single inference pipeline.
VisionVitals is a health monitoring application that analyses a subject's face and hand from video input using computer vision models to measure vital signs. The engineering challenge was integrating a GPU-accelerated PyTorch inference model into a web application without blocking request threads or coupling the ML environment to the Django process.
The architecture separates concerns into three layers: a Django + DRF web API that accepts uploads and returns results; a Celery task queue that dispatches inference jobs; and a FastAPI microservice that hosts the MediaPipe pose estimation and PyTorch classification pipeline and can be scaled or redeployed independently of the web tier.
Built as part of an Omdena collaborative project with a distributed international team.
Outcome
End-to-end AI inference pipeline: Django REST API receives video frames, Celery queues inference tasks, FastAPI microservice runs MediaPipe pose estimation and PyTorch classification, results returned asynchronously. Architecture decouples ML inference from the web layer for independent scaling.