Dev turned Operator. I bridge the gap between "AI Hype" and "Production Reality."
I don't just advise — I build the engine for value creation. Ship production LLM platforms (0→1 fast), enable engineering teams, and treat AI as infrastructure, not a feature.
Now: Operating Principal (AI) @ GrowthFactors (Bregal Unternehmerkapital) · Founder @ DocIQ Prev: Head of AI @ CISTEC AG
| 🔨 Builder | Building DocIQ — AI document intelligence for legal. Word-native tracked changes, legal research across 6 databases, court anonymization with zero persistence. Swiss engineered. |
| ⚙️ Operator | Operating Principal (AI) @ GrowthFactors — deploying a standardized AI operating system across Bregal's portfolio companies |
| 🎓 Technical Expert (ITDR) | Listed expert for Swiss arbitration & complex data disputes |
| 💡 Advisor | Guide VCs, C-suites, and portfolio companies on AI strategy, architecture, and risk via Mickel Tech |
Creating robust, production-ready AI systems that drive measurable outcomes
🏗️ Reusable AI Platforms
→ LLM API gateways, vector search services, context engineering, evaluation frameworks
📄 AI Document Intelligence (DocIQ)
→ OOXML-level editing with tracked changes, legal research across 6 databases,
court anonymization with zero data persistence, agentic playbooks
🔒 Privacy-Preserving AI
→ On-prem PII masking (~97% accuracy) for FADP/GDPR compliance
🤖 Agentic Architectures
→ Multi-agent systems, prompt-to-structured-data compilers, semantic search at scale
👨💻 Developer Enablement
→ AI coding assistant rollouts, prompt/playbook patterns, safety guardrails
📊 Production LLM Ops
→ Evaluation pipelines (quality/safety/latency/cost), observability, A/B testing
| 📄 | Shipped DocIQ Shield (court anonymization, in production at Swiss courts) and DocIQ Sphere (the most powerful agentic legal document platform: 40+ tools, 6 legal databases, Word-native tracked changes) |
| ⚡ | Rolled out AI-native SDLC across 4 dev teams on heterogeneous stacks — 100K+ word engineering guide, predicted 50%+ faster delivery |
| 🎙️ | Architected production AI voice agent for enterprise call center — eval framework, prompt optimization, real-time API |
| 📊 | Built gmickel-bench — real-world AI coding evals across 6 benchmarks with dual scoring (LLM judge + human review) |
| 🏥 | Shipped a multi-tenant clinical LLM platform in <3 months with on-prem PII masking (97% accuracy), driving new ARR |
| 🎤 | Speaker at openEHR.ch Symposium: Agentic AI for Structured Clinical Data |
- 🧭 Flow-Next - Spec-driven agentic SDLC for coding agents: durable specs keep intent alive across the chat, context-fit plans size each task to one iteration, fresh-context workers re-anchor per task, and adversarial cross-model review gates (RepoPrompt / Codex / Copilot) run until SHIP — with a receipt for every handoff. Team-ready: handover objects, R-ID requirement coverage, make-pr cognitive-aid PRs, live-app QA, and two-way Linear / GitHub Issues sync. Ralph Mode for full autonomy; first-class on Claude Code, Codex, Droid, Grok Build, and Cursor. The agent loop Anthropic's own guidance recommends but their plugin doesn't implement.
- 🧪 flow-next-opencode - OpenCode port of Flow-Next. Same re-anchoring, same review gates, now with OpenCode Agent as reviewer. Experimental.
- 🧠 GNO - Local search, retrieval, and synthesis for the files you actually work in. Hybrid BM25+vector retrieval, grounded answers with citations, wiki links, backlinks, knowledge graph. CLI, MCP, Web UI, REST API, SDK — for humans and AI agents. Publish any note to gno.sh as a polished reader URL (public, secret, invite-only, or locally-encrypted-before-upload). Markdown, code, PDF, Office — 100% local by default.
- 📊 sheets-cli - Google Sheets primitives for humans and agents. Pipe-friendly.
- 🔖 raindrop-skill - Agent Skill for Raindrop.io bookmarks. Works with Claude Code, Codex, Amp, OpenCode.
- 🎓 better-skill-builder - Teaches AI agents to build skills. Battle-tested patterns, not theory.
- 📧 outlookctl - When Graph API is blocked, COM whispers to Classic Outlook.
- ⚙️ claude-code-config - My Claude Code setup. Steal freely.
- ✂️ SmartTrim - AI clipboard carnage? Fixed. Ghost indentation, begone.
- 🏄 cursor-windsurf-convert - Surf your AI rules between Cursor and Windsurf. Lossless.
- 📅 CalSync - Clone Apple Calendar events. Simple utility, does one thing.
- 🤖 CodeWhisper - The OG plan→implement agent. Still my daily driver for repo context.
- 🧠 memorybot - Unlimited context chatbot. Before it was cool.
- 🏗️ turborepo-shadcn-nextjs - Next.js + Nextra + Storybook + shadcn/ui. Bun & Biome powered.
- 🎮 CodeQuest - Quiz game framework. Build your own learning adventures.
- 🏥 openehr-quest - Learn OpenEHR via CodeQuest
- 💡 hugh - Node.js Hue bridge lib
- 🤖 telegram-hue-bot - Lights via Telegram
| Focus | What It Means |
|---|---|
| 🔄 Rewire SDLC | Move teams from Copilot autocomplete to AI-native workflows (Plan → Build → Deploy) |
| 🤖 Ops Agents | Replace manual back-office loops (Finance, HR, Support) with autonomous systems |
| 📈 Product AI | Ship features that drive NRR & EBITDA — 30-day lighthouse pilots, kill or scale |
| 📅 | Advisory/consulting: Book a call |
| 💬 | Questions/feedback: DM on X |
| 🚀 | Open to select opportunities where technical depth meets strategic impact |
| ❤️ | Support: GitHub Sponsors · Ko-fi |
- The "Context Hygiene" Problem: Why I rewrote my Claude Code workflows
- Announcing gmickel-bench: Real-World Evals
- Claude Opus 4.5 Unlocks the "No Restart" Workflow
- The Merchants of Complexity: Why AI Finally Delivers What Agile Promised
➡️ More posts on Medium · More on Substack ⬅️
If my work helped you, consider supporting it:
Built with ❤️ by Gordon Mickel