Software that holds up under audit.

I build systems where correctness matters and where the rules change often. Tax law and financial reporting are the hardest version of that problem, so that is where most of my experience sits. If your team is running a critical process out of a spreadsheet held together by one person, that is a good first conversation.

• Financial systems and reporting tools
• Tax calculation and filing workflows
• Internal tools that replace spreadsheets
• APIs, integrations, and data pipelines

Most AI projects that fail were never a model problem. They were a workflow problem with a model bolted on. I start by finding out what the process actually does today, then work out whether a model belongs in it at all. When one does, the work is mostly in the parts nobody demos: evaluation, cost control, and what happens when the model is wrong.

• Workflow review before any model gets picked
• Document extraction and classification
• Retrieval over your own documents
• Evaluation, cost limits, and fallbacks

Games are where I test ideas that would be too risky in a regulated product. The constraints are familiar: hard performance budgets, real-time state, and players who will find every edge case you did not. I take on gameplay and systems work, internal tooling, and prototypes that exist to answer one specific question before you commit a year to it.

• Gameplay and systems programming
• Tools and pipelines for small teams
• Prototypes built to answer one question
• Shipping and post-launch support