AI products. Data platforms. Web experiences. From the first messy idea to the version people actually use. No six-month discovery phase. No 87-slide strategy deck. We get into the build.
Currently building: AI agents / data systems / questionable side experiments
Four places where we do our best work. Open one to see what is inside.
AI agents. RAG. Automation. Analytics. Pipelines. Less dashboard theatre, more systems that think and move.
Fast products. Clean architecture. Interfaces people don't need a tutorial to understand.
We design interfaces that feel obvious before anyone explains them, drawn beside the engineers who build them.
Experiments. Internal tools. Prototypes. Weird ideas. Some become products, some become lessons, and some probably shouldn't have left localhost.
The first commit. The ugly prototype. The thing that broke production. The fix that finally worked. See what happened between idea and shipped.
Four systems, from schema discovery to production billing. Open one to see how it was built, and what it actually looks like.
DataNexusSchema discovery and connector validation, before a row moves.
A control plane that discovers schemas, validates connectors and provisions multi-tenant ingestion, so onboarding a new operational database is a review rather than a fortnight of manual setup.
FlowStackPostgres to warehouse, with the validation in front of it.
Schema, null and type validation applied at the boundary, plan normalisation behind it, and ETL orchestration that deploys from one command instead of a runbook nobody has read since January.
Carrier billingBilling workflows used to bounce between people, spreadsheets and manual approvals.
We turned the whole thing into one automated system. Payloads land, anomalies are caught at the schema boundary, and Bronze-to-Gold billing data comes out the other side with line-item reconciliation and monitoring in front of it.
141successful workflows
Ghazanfar Sheikh, DevVaults, measured 2026-09-14. Counted by the studio from the workflow runs completed by this system, and attested rather than exported from a monitoring tool.
Enterprise AIForecasting, anomaly alerting and query in plain language.
Predictive forecasting and anomaly detection over the warehouse, with a natural language query panel in front of both, so the question gets asked by the person who has it.
It shipped. It stayed shipped.
Field notes
Four scenes from the studio. The hardest problem goes on the table early, the architecture gets argued over and redrawn, the talking stops, and the unglamorous checks happen before you ever see a demo.
Not your title. Not your seniority. The idea. Working sessions exist to expose the problem while it is still cheap to be wrong about it.
Engineering × Product
Engineers beside designers. Designers beside product. Nobody throwing Figma files over walls.
Working session
Headphones on. Slack ignored. The debate is over and somebody has to make the thing work.
Build · Test · Repeat
The 8 pixels nobody requested. The edge case nobody noticed. The interaction that simply feels better.
Review · Validate · ShipWe don't hire people to stay in their box. Curious people make better products. See what's open→
The one sitting in your Notes app. The one everyone says is ambitious. The one that might actually work. Send it our way.