Build an Inference API for a Cancer Detection Image Model (Research Prototype)

We’re researching AI-assisted cancer detection for hospital pathology workflows. We have an image classification model trained on the public PatchCamelyon (PCam) dataset that classifies tissue patches as cancerous or healthy, and we need a backend service that makes it easy for other tools to send images and get predictions back. You’ll build the inference API, package it for deployment, and document it so our team and partner applications can integrate with it. This is a research prototype for internal evaluation. All testing uses public PCam images. Scope: POST /predict endpoint that accepts an image upload and returns the predicted label, confidence score, and model version POST /predict/batch endpoint for up to 50 images in one request POST /explain endpoint that returns a Grad-CAM heatmap overlay showing which regions drove the prediction GET /health and GET /model-info endpoints (model name, version, input size, decision threshold) Input validation for file type, size, and image dimensions, with clear error messages API key authentication and basic rate limiting Request logging (timestamp, latency, prediction, model version) with image data kept out of the logs Configurable decision threshold via environment variable Deliverables: Source code in a private GitHub repo (Python with FastAPI, PyTorch or ONNX Runtime for inference) Dockerfile and docker-compose setup that runs the service locally with one command, on CPU, with optional GPU support OpenAPI documentation for all endpoints, with example requests and responses Automated tests covering valid images, invalid files, batch requests, auth failures, and threshold changes A simple benchmark script reporting average latency and throughput for single and batch requests README covering setup, swapping in a new model checkpoint, environment variables, and deployment notes A short screen recording showing a prediction and a heatmap returned end to end Requirements: Strong backend experience with Python and FastAPI (or a similar framework) Hands-on experience serving deep learning models in production (PyTorch, TorchServe, ONNX, or similar) Solid API design: validation, auth, error handling, versioning Experience with Docker Nice to have: Experience with medical imaging or healthcare APIs Familiarity with model explainability tools such as Grad-CAM Awareness of privacy practices for health data (de-identification, audit logging)