I get frontier models into production in places where the regulatory and security bar is high. Anyone can demo an agent. The hard part is the substrate underneath: review gates, observability, permission boundaries, and a hard rule that AI never touches the system of record.
Some things I've built:
- The AI delivery function at a regulated commercial lender, from zero: a skill marketplace, a publish-and-audit governance gate, drafts-only by default, cost-per-skill observability. The first time one skill ran in front of the team, it was about to ask a broker for a borrower's SSN. What we changed after that is the interesting part: davidveksler.com/david/ai-strategy.html
- Agentic engineering at enterprise scale at Antech. Claude Code and Copilot workflows that built a support platform end to end and automated unit testing, a 4.2x velocity increase.
- A decade in regulated fintech and digital assets before that: CCSS Level 3 custody, wrapped tokens at $100M+ TVL, and recovery tooling that scanned vaults across 20+ EVM networks to bring back $4M+ in stranded assets.
~20 years of engineering, principal/staff level. C#/.NET, Python, TypeScript/React, Azure, AWS, SQL, OpenTelemetry. I reach for frontier models plus retrieval before fine-tuning, almost every time.
I don't train models and I'm not a prompt engineer. I'm the person you bring in when the demo worked and now it has to survive compliance.
Reach me: [email protected] · LinkedIn · Substack