ML Deployment Platform
Unified control plane for deploying and monitoring ML models across AWS, GCP, and Azure without touching three cloud consoles.
Role
Full Stack Developer
Timeline
2024 — 2025
Stack
Overview
A comprehensive ML deployment control plane for managing machine learning model endpoints across AWS, GCP, and Azure. The platform enables teams to deploy, monitor, and manage ML models without direct cloud console access through an intuitive web interface. Features include asynchronous job processing, real-time status updates, deployment log streaming, and automated infrastructure orchestration using Pulumi and Ansible.
Architecture & Decisions
Multi-Cloud ML Infrastructure Management
Provides unified interface for deploying ML models across AWS SageMaker, Google Cloud Vertex AI, and Azure ML. Features include automated EC2 provisioning with custom AMIs, security group configuration, Ansible-based instance setup, and support for both CPU and GPU instances. Handles model lifecycle operations including deploy, start, stop, restart, and terminate with full cost tracking.
Real-Time Operations & Observability
Built with asynchronous job processing using BullMQ and Redis for non-blocking operations. Implements Socket.IO for real-time status updates, deployment log streaming, and progress tracking. Features NextAuth authentication with email/password and GitHub OAuth, secure webhook callbacks, and centralized logging with failure handling. Enables teams to manage production ML deployments with enterprise-grade reliability.
The Hard Part
Deployment operations that touch real cloud infrastructure can't block the UI or fail silently. The hard part was making every deploy/start/stop operation asynchronous and observable in real time, across three very different cloud APIs.
Outcomes
Deploy speed
10x faster
Clouds
AWS · GCP · Azure
- —Asynchronous job processing with BullMQ + Redis and automatic retries
- —Pulumi Automation API for programmatic infrastructure provisioning
- —Socket.IO streams live deployment logs and status updates
- —NextAuth authentication with email/password and GitHub OAuth
Up Next
View allMulti-Cloud AI Deployment
Python tool that deploys open-source models (Llama 3, BERT, GPT variants) across AWS, GCP, and Azure through one unified API.
Read case study