rl-eval-k3s
A containerised ML evaluation service with FastAPI and Kubernetes (k3s) manifests.
Pythonopen source
Problem
Evaluating trained policies repeatably is its own engineering problem. The aim was to make evaluation a proper service you can deploy and scale, rather than a notebook someone runs locally.
Approach
- Wrapped policy evaluation in a FastAPI service with a clean HTTP interface.
- Containerised the service with Docker for reproducible runs.
- Provided Kubernetes (k3s) manifests so evaluations can be scheduled and scaled on a lightweight cluster.
- Structured it as an MLOps reference for serving evaluation workloads.
Tech
FastAPIDockerKubernetesk3sPython
Status
Shipped and open source. Demonstrates the MLOps side of my RL work.