Every project, in order.
Seven case studies across games, spatial computing, ecommerce AI and finance. The dates are the argument.
Agentic search over 10,000-page filings
The answer is a number in a table, not a sentence you can paraphrase. Wealth managers work from filings most models won't even open. Built the retrieval layer that does, and made it abstain rather than guess.
Neoflo: AI for the CFO tech stack
Finance teams are buried in invoices, reconciliations and expense approvals. Neoflo runs those workflows end to end instead of summarising them. Live with enterprise clients, processing 6,000+ invoices a month across six countries.
Neoflo IDP: intelligent document processing
We benchmarked six platforms before building anything. Three of the five we could buy fell apart on handwritten invoices. So we built the reader ourselves. The gap was never about optimisation. It was about approach.
Soul: my digital twin
Six AI tools, six silos. Every session started from zero. So I built the memory layer they were all missing. Now whichever tool I open already knows what I've decided this week.
Agentic AI Platform
Most AI deployments fail at the seams. Between agents, between tools, between a person and a machine. So this one holds there. Five agent layers. A confidence floor that's stricter in production than in testing. And a bad call that can't spread past the step that made it.
Velocity: AI video generation for enterprise catalogues
HP needed 2,500 product videos. Agencies quoted $1.25M and 6 months. We shipped it in 7 minutes per video, at $50 each, across 28 markets.
AN-Prez: multiplayer holographic presentations
Mixed reality presentations on HoloLens with shared holograms and spatial annotations. Enterprise deployments, shipped on the Microsoft Store, spatial computing before anyone called it a category.
