Evan S.

My Claude agent has run a real business's support inbox unattended since July. 877 threads, 97% resolved without me. It looks up customers in the database, replies, ships small code fixes, and hands off what it shouldn't decide alone. That's the work I do for clients, in three shapes: Build. Claude agents, skills and automations that run in production, with the guardrails that keep them from going quietly wrong. Audit. An outside look at the AI systems or automations you already have. What's fragile, what's missing, what to fix first. Train. Getting your team using Claude Code and shared skills consistently, not just the one person who's good at it. Recent work: — Multi-tenant therapeutic AI platform, 657 files, built solo with Claude Code. 14-step pipeline, HIPAA-compliant. — AI chatbots for a 3M MAU entertainment platform (Claude + GPT-4) — Coaching analytics platform used by 50%+ of NBA teams — Fractional CTO for a 60-person, 3-entity company: tech stack audit, Zapier to n8n, handed over with docs — goodgame.md, an open-source behavioral tuning framework for Claude Code Before this, I scaled a Web3 agency from zero to $20M in two years and turned a 40-year-old special education publisher into a digital SaaS platform serving 100K schools and 1M students. Claude API, Claude Code, MCP, RAG, n8n, React, TypeScript, Node, Laravel, Python, PostgreSQL, AWS, Azure. 15 years. If you're running LLMs in production and something feels off, I've probably seen it.