Get AI right before you build it. I help teams decide where AI belongs in their product, which models to use, how to know it works, and how to develop with AI agents without losing control of the codebase.
From "we should do something with AI" to a feature your users actually rely on. We work out where AI adds real value in your product, design how it behaves when the model is wrong, and plan how it gets from prototype to production.
The biggest model is rarely the right answer. I help you compare models on your actual tasks and weigh quality against cost, latency and privacy, so each job runs on the model that fits it.
Prompts and models change constantly. A testbench runs your real cases against every change and shows what got better and what broke, before your users notice. I help you build one that fits your product and your team.
AI coding agents like Claude Code can make a team much faster, or produce a lot of code nobody understands. I help you structure your codebase so agents work reliably in it, and coach your developers on workflows that keep both the speed and the quality.
I wrote get_it, one of the most used Flutter packages, and the flutter_it family around it, and I have written and spoken about app architecture for years. Agents need the same thing people do: a clear structure where every piece has one obvious place.
Every app I have built since then was developed with AI agents. LatinaUGC, a two-sided marketplace with WhatsApp and payment integration, took about two months. The conventions, guardrails and review habits I recommend are the ones I use myself every day.
I have seen many hypes come and go. That helps me separate what AI really changes from what it does not, and keep the fundamentals in place while a team adopts new tools.
Many engagements start with an assessment, so we both know exactly where you stand. If you already know where you are headed, we can also start directly with coaching. Both can be booked on their own.
A focused review of your codebase, backend and plans. You get a written report and a concrete roadmap, discussed together in a handover call.
Guidance while your team does the work. Blocks are used for scheduled video sessions and for asynchronous code and architecture reviews.
When every question means a new quote and a new approval, teams hesitate to ask, and small questions quietly grow into expensive problems. With a block already booked, asking is easy: your team reaches out as soon as something is unclear, while it is still cheap to fix. Short questions between sessions are included.
An example: a team wants to move its app from FlutterFlow to a conventional Flutter codebase that its own developer maintains with Claude Code. The assessment compares refactoring the generated code with a clean rebuild and designs an architecture the agent can work in safely. The coaching then accompanies the developer through the migration.
Your team. My role is advisory: architecture, methodology, reviews and coaching. You own all code and repositories from day one. If you want hands-on help with specific parts, for example complex components or AI features, I can quote that separately as contract work.
No. Model selection, testbenches and AI workflows apply to any stack. Flutter is where I can also help with the app itself.
No. I recommend what fits your requirements, whether that is a hosted model from one of the large providers or an open model you run yourself.
Yes. A review of an existing feature, its prompts and how it is tested is often the fastest way to find out where it breaks and what to improve.
I work remotely, from Germany, Colombia or some other sunny place. Wherever I am, we will find regular slots that work for teams in Europe and the US.
All project materials and business information stay confidential. A mutual NDA can be signed on request.