AI that survives contact with production.
An independent engineering practice for banking, fintech, and legal — where reliability, privacy, and compliance are not optional. Built on EU infrastructure. Kept honest by evals.
Five kinds of work.
AI features in your product
Search that understands intent, drafts that match your tone — built into your app, with the evals to keep them honest.
Internal tools that compound
Document review, support triage, data extraction — the systems your team uses daily, made dramatically faster.
From-scratch AI products
Architecture through launch for AI-native founders, with the evaluation system that survives real users.
AI infrastructure & DevOps
Kubernetes, Terraform, GitOps, model serving, observability — proven in regulated banking environments.
Strategy & advisory
Due diligence, architecture review, fractional CTO — clarity on what to build before anyone builds it.
Not sure which?
A two-week diagnose sprint decides it. About a third of intro calls end with "don't build this" — that honesty can save you a quarter.
Two products, built end to end.
Portfolio analytics built for crisis scenarios.
Risk decomposition, factor analysis, stress testing — with conversational AI agents so a fund manager asks plain-English questions and gets rigorous answers back. Built with Raphaël Douady's risk mathematics.
demo.radomir.fr →A lawyer's AI assistant, connected to their actual case files.
Connects the AI tool a lawyer already uses — Claude, ChatGPT, Cowork — to their case files. Drafts pleadings, verifies quotations character-for-character, traces every citation to a bundle page. Validated on a real UK Court of Appeal matter.
The model is the easy part. The lasting work is the evaluation harness that keeps it honest after we're gone.
One senior engineer, end to end.
Magenta Code is the practice of Radomir Klacza — fifteen years of platform engineering in banking and regulated finance, from Linux at Citi to running an Ethereum staking platform on Azure Kubernetes at SIX Digital Exchange.
Today he builds AI risk analytics at Fundrank alongside Raphaël Douady, Research Professor of mathematical finance at the Sorbonne.
The person on the discovery call is the person who writes the code — and designs the eval, and stands behind the result.
Thirty minutes. Just your problem.
Tell us what you're trying to build. If we can help, we'll say so. If we can't, we'll point you to someone who can.