AI systems builder, transformation operator, and independent advisor
I build production AI products and the operating systems around them. My portfolio includes Flow-Next, GNO, Dettivo, and DocIQ; my advisory work helps software companies redesign product delivery and put AI systems into production.
Now: Operating Principal, AI & Technology at Growth Factors (Bregal Unternehmerkapital); founder of DocIQ and Mickel Tech
Previously: Head of AI at CISTEC AG
| Builder | Build and operate a portfolio of AI products: agentic software delivery with Flow-Next, local knowledge infrastructure with GNO, private speech tooling with Dettivo, and regulated legal AI with DocIQ. |
| Operator | Operating Principal, AI & Technology at Growth Factors. Lead portfolio-scale AI transformation across Bregal software companies, from executive alignment and operating-model design through hands-on product and engineering rollout. |
| Advisor | Through Mickel Tech, take selected independent mandates in agentic PDLC, production AI systems, architecture, build-vs-buy decisions, and internal capability building. |
| Technical expert | ITDR-listed independent expert for complex software disputes, technical due diligence, and contract acceptance. A specialist practice alongside the product and transformation work. |
| ⚡ | Designed and rolled out an AI-native product and software delivery methodology across 10+ enterprise software companies, reaching 700+ R&D engineers through PDLC redesign, change management, hands-on coaching, and the open-source Flow-Next delivery system |
| 📄 | Shipped DocIQ Shield, court anonymization with zero data persistence, and DocIQ Sphere, a Word-native legal agent with 40+ tools and research across six legal databases |
| 🧠 | Built a multi-tenant intelligence platform that turns noisy operational signals into validated, role-aware executive insight; live across three companies, with a patent-pending measurement methodology |
| 🎙️ | Advised on enterprise voice, call-analysis, and agent-coaching systems operating at roughly one million calls per month |
| 🏥 | Shipped a multi-tenant clinical LLM platform in under three months, including on-premise PII masking at approximately 97% accuracy |
- 🧭 Flow-Next - Repeatable agentic engineering. The workflow layer that turns AI coding agents into a disciplined delivery system: durable specs preserve intent, fresh-context workers prevent drift, adversarial cross-model reviews fix issues until
SHIP, and receipts prove every handoff. Everything lives in your repo, with zero external dependencies. Run it human-in-the-loop or hand a reviewed spec to Ralph overnight. Claude Code · Codex · Cursor · Droid. flow-next.dev - 🧪 flow-next-opencode - OpenCode port of Flow-Next with Ralph mode, re-anchoring, and multi-model review gates.
- 🧠 GNO - Local-first search and retrieval for working files. Hybrid BM25, vectors, reranking, wiki links, backlinks, and graph traversal via CLI, MCP, REST API, SDK, and Web UI. Markdown, code, PDF, and Office documents stay local by default. gno.sh
- 🎙️ Dettivo - Private speech workstation for Apple Silicon. Local dictation and meeting transcription with WhisperKit, Parakeet v3, and a custom MLX model; CLI, REST API, and MCP access.
- 📄 DocIQ - Agentic legal document intelligence. Word-native tracked changes, research across six legal databases, reusable playbooks, data rooms, and zero-persistence court anonymization.
- 📊 sheets-cli - Composable Google Sheets CLI and agent skills for Claude Code and Codex.
- 🗂️ obsidian-skill - Agent skill for reading, searching, and maintaining Obsidian vaults through the official CLI.
- 🎓 better-skill-builder - Production patterns for building portable Agent Skills.
- 🔖 raindrop-skill - Bookmark search and management for Claude Code, Codex, Amp, and OpenCode.
- 📧 outlookctl - Local automation for Classic Outlook when Microsoft Graph is unavailable.
- 📈 gmickel-bench - Real-world coding-agent evaluations with model judging and human review.
- ✂️ SmartTrim - macOS menu-bar utility that repairs mangled text copied from AI coding assistants.
- ⚙️ claude-code-config - My working Claude Code configuration and reusable setup patterns.
- 🏄 cursor-windsurf-convert - Lossless conversion between Cursor and Windsurf rule formats.
- 🤖 CodeWhisper - The 2024 predecessor to Flow-Next: end-to-end task implementation and fast codebase-to-LLM context assembly.
- 🧠 memorybot - Persistent-context Node.js AI agent, first released in 2023.
- 🏗️ turborepo-shadcn-nextjs - Bun and Biome-powered monorepo starter.
- 🎮 CodeQuest - Framework for building interactive learning games.
- 🏥 openehr-quest - OpenEHR learning game built on CodeQuest.
- 💡 hugh - Node.js library for Philips Hue bridges and lights.
- 🤖 telegram-hue-bot - Telegram control for Philips Hue.
- 📅 CalSync - Archived macOS utility for copying events between Apple calendars.
- AI-native PDLC: connect product discovery, specification, implementation, review, QA, and measurement into one agent-ready system
- Agent reliability: make autonomous work inspectable through evidence, adversarial review, and durable handoffs
- Private AI: keep sensitive legal, clinical, and enterprise data under the owner's control
- Compounding tools: turn repeated delivery problems into open-source infrastructure other teams can adopt
I write field notes on agentic engineering, context design, evaluation, and production AI at mickel.tech. The older archive lives at Byte-Sized Brainwaves.
- Harness Engineering: A Field Guide to Building Agents That Hold Up
- Designing Tools AI Agents Can Actually Use
- Agent Memory Is Not a Vector Database
- Evaluating Agents When the Demo Always Passes
- When to Reach for Multiple Agents (and When Not To)
- Advisory and consulting: book a call
- Professional profile: LinkedIn
- Open-source questions: GitHub or Discord
- Email: gordon@mickel.tech
Basel, Switzerland