The model writes code.I still own the outcome.
Where it genuinely helps, where it does not, and what I do to stop a confident wrong answer reaching a user.
Where it genuinely helps, where it does not, and what I do to stop a confident wrong answer reaching a user.
Ordered roughly by how much time each one gives back. None of it replaces reading the diff.
Scaffolding, migrations, release notes, test fixtures. The repetitive half of the job, off my plate.
A second pass over diffs before they reach a teammate — dead paths, missed edge cases, naming that has drifted.
Breaking down requirements, mapping data flows and challenging assumptions before the first migration is written.
React, Laravel, TypeScript. Shortens the gap between a ticket and a reviewable pull request.
Working slices instead of mockups — validated with real data before engineering time is committed.
Assisted validation, classification and summarisation wired into real workflows, with a human path for the edges.
Claude, Cursor and custom MCP servers sharing context in one loop instead of working in isolation.
It isn't vibe coding. It's systems thinking, architecture and the part where it has to keep running.
Each one carries live traffic. Each one has a path for when the model gets it wrong, because it does.
National media contest engine with automated drawing logic, cron-driven workflows, and AI-assisted coupon validation. Tuned for spikes from broadcast and print campaigns.
Read the case studyMental wellness SaaS with AI-driven personalization, GraphQL APIs, in-app purchases, and FCM-powered engagement notifications. Shipped a complete MVP from architecture through deployment.
Read the case studyDocument management and approval workflow platform — digital onboarding, identity/compliance verification, and AI automation. Webhooks and APIs for CRM integration.
Read the case studyA short list on purpose. These are the ones I know well enough to be fast in.