AI-enabled Software Engineer

A model running in a notebook or a local script is not a production system. The gap between "it works on my machine" and "it serves requests reliably, restarts cleanly, and takes updates without downtime" is a software engineering gap, not a modelling one. It calls for API design, containerisation, CI/CD, and infrastructure as code, skills that many ML-focused teams simply don't have in-house.

I build most AI systems around a FastAPI backend, chosen for being async native, supporting streaming responses out of the box, generating interactive API documentation automatically, and running with low enough overhead for real-time inference and batch endpoints alike. Every service ships with a production-grade Dockerfile using multi-stage builds, a non-root user, and pinned dependencies. Images are pushed to Amazon ECR and deployed to AWS EC2 behind Nginx, with S3 handling model artefacts and file storage through pre-signed URLs where needed. GitHub Actions runs the CI/CD layer end to end: lint, test, build the image, push to ECR, and deploy to EC2 on merge, with no manual step in between. On the ML side, MLflow tracks experiments and DVC versions data, so every deployed model can be traced back to the exact code and data that produced it.

A containerised FastAPI service running behind a fully automated deployment pipeline, with monitoring hooks and reproducible builds built in from the start. SmartInsureAI runs on this exact stack: data pipelines, a FastAPI prediction API, and a CI/CD path through GitHub Actions, Docker, ECR, and EC2 that ships model and code updates automatically on every merge.