Fast-charging stations for electric vehicles across Europe. I manage two engineering teams.

• Charging: the OCPP layer, 99.9% uptime across 2.9k+ chargers.
• Roaming: logic and charge cards that process 80-90% of Fastned’s EUR 122.4M 2025 revenue. We don’t generate it, we make sure it clears.
• Restructured the original Protocols group into these two autonomous teams.
• Led architecture and delivery for Fastned’s biggest transformation: charge point operator to mobility service provider.
• Founded and lead the AI Squad, plus a suite of internal apps: workshops, dashboards, and cost-optimized usage that cut delivery effort by an order of magnitude.

Most of the job is keeping revenue-critical infrastructure boring, which takes more work than it sounds.

Agentic analytics: ask a business question in plain English, get a live dashboard back.

• Co-founded and ran it as CEO: product, architecture, data modeling, plus sales, demos, contract negotiation.
• Platform ran in production with multiple clients.
• A distribution partnership turned into a consultancy relationship across their 300 clients.
• Jan 2026: the big AI labs shipped the core of what we built into their own products. Platform path closed.
• Pivoted to a data and AI consultancy: MCP layers and pipeline structure for legacy SaaS. Winding down now.

When your core value sits on a foundation lab’s roadmap, every model release is an existential event. I can confirm this from experience.

Virtual try-on for fashion e-commerce.

• Rebuilt the product and technical org from scratch: architecture, infrastructure, execution, client relationships.
• Empty repo to a production system with paying customers in under three months.
• EUR 300K ARR by month three.
• Stepped back in Nov 2024 for geography reasons, moved to Fastned.
• Still running today, run by my cofounders without me. That’s the outcome I’d pick again.

Founding engineer at a fintech startup, four and a half years, one of the first hires.

• Wrote across the stack: C# and Java tools for internal and customer use, Scala and Python for data work and debugging, Bash-driven automation for the build pipeline.
• Up to 200% throughput and 80% faster deployments from the infrastructure work.
• Led the company’s largest client integration end to end: requirements, product changes, fast issue resolution.
• Recruited and onboarded engineers as the team scaled.

Five years watching a startup grow from a handful of people into a real company. A different kind of education than the one before it.

General speech engines choke on specialty vocabulary. This one didn’t.

• Built for an NHS client to transcribe medical dictation across Haematology, Urology, Radiology, and Oncology, each with its own terminology.
• Trained the model on YouTube lecture and conference audio with matching transcripts instead of hand-labeling data from scratch.
• Team project at UCL.

The client-specific vocabulary problem in speech recognition never went away. It just moved to LLMs.

Most of my code these days is written by coding agents.

• The hard part isn’t prompting, it’s operations. Agents are stateless, so everything durable has to live in files.
• Fixed root files, one job each, read in the same order every session.
• Hard line caps so state always fits a fresh context window.
• Repeated tasks become executable skills instead of being re-improvised.
• Every handoff gets written down; the next agent reads it first.
• Verified like code: spawn a fresh agent, quiz it against ground truth, fix the files, not the agent, when it fails.