// the find
ebhy/budgetml
Deploy a ML inference service on a budget in less than 10 lines of code.
BudgetML wraps GCP preemptible instances with FastAPI, SSL, and OAuth2 so a data scientist can get an ML model behind a real HTTPS endpoint without touching Docker, Nginx, or GCP directly. It targets the gap between 'just use Flask locally' and 'stand up a Kubernetes cluster' — useful for demos, side projects, and low-traffic internal tools. The catch: it's explicitly abandoned, with the README asking for a new maintainer.
Preemptible instance + auto-restart via Cloud Scheduler and Pub/Sub is genuinely clever — you get ~80% cost reduction with only a few minutes of daily downtime. SSL via docker-swag/LetsEncrypt is handled completely automatically, which is the part that usually bites data scientists. The predictor interface is clean: implement load() and predict(), and you're done. FastAPI over Flask means async support and free Swagger UI out of the box.
Abandoned since 2024 with no active maintainer — GCP API changes or Python version bumps will silently break things and nobody is fixing them. Preemptible (now Spot) VMs get killed with 30 seconds notice; the autorestart via a Cloud Function adds complexity and still means your endpoint is down for 3–5 minutes daily, which is a problem if anyone actually depends on it. Hard-wired to GCP only — no AWS, no Azure, no bare metal. The authentication story is basic HTTP Bearer; if you need anything beyond a single shared password, you're on your own.