Fusion: Architecting Resilient Container Setup
Architected a recoverable container workflow for an AI coding product.
- State Machines
- Container Orchestration
- Developer Productivity
- AI Coding Agents
- Production Reliability
Midhun Darvin
Software Engineer at Builder.io
From their time as
Software Engineer
Builder.io • 2024 - 2026
Overview
Midhun architected a state-machine approach for Fusion's container setup flow. It handled provisioning, repository access, dependency installation, retries, and faster paths for already-running environments.
The Story
Midhun worked on Fusion, a Builder.io product that lets non-technical users request website changes through an AI coding workflow and see the result in a live visual environment.
The environment setup involved several dependent steps: provisioning a container, cloning a repository, installing dependencies, starting a dev server, and connecting the interface to the coding agent. Capacity limits, missing storage, and expired GitHub credentials could interrupt the flow.
He modeled the workflow as a state machine rather than a sequential process. That made the paths, forks, retries, and recovery states explicit, while allowing already-running environments to take a faster route.
He also owned parts of the early architecture, WebSocket connection, code-diff experience, deployment work, and the feedback loop that translated customer issues into product changes.
Ownership Snapshot
Broad role
Software engineer on Builder.io's Fusion product.
Goal
Make AI-driven website changes visible and reliable for users.
Direct ownership
Owned parts of the early architecture, WebSocket flow, and production feedback loop.
Team execution
Worked with product and engineering to deliver and monitor the workflow.