I build tools for agentic engineering and put AI systems into production.
I build Flow-Next, GNO, Dettivo, and DocIQ. At Growth Factors, I lead AI and technology value creation across BU's European portfolio of more than 30 companies, working directly with product and engineering teams.
Now: Building Flow-Next, GNO, Dettivo, and DocIQ; Operating Principal, AI & Technology at Growth Factors (BU Bregal Unternehmerkapital)
Previously: Head of AI at CISTEC AG
| Builder | I build and operate Flow-Next, GNO, Dettivo, and DocIQ. They cover agentic engineering, local knowledge infrastructure, private speech tooling, and regulated legal AI. |
| Operator | At Growth Factors, I lead AI and technology value creation across BU's European portfolio of 30+ companies. The work happens inside portfolio companies, with their product and engineering teams, and often inside their codebases. |
| Advisor | Through Mickel Tech, I take selected independent mandates in agentic engineering, production AI systems, architecture, build-vs-buy decisions, and internal capability building. |
| Technical expert | I am an ITDR-listed independent expert for complex software disputes, technical due diligence, and contract acceptance. This is a specialist practice alongside my product and operating work. |
| โก | Built and rolled out an agentic engineering programme across 10+ enterprise software companies, working directly with product and engineering teams and supporting the rollout with the open-source Flow-Next system. The programme reaches 700+ R&D engineers |
| ๐ | 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 |
| ๐ง | 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 |
| ๐๏ธ | Advised on enterprise voice, call-analysis, and agent-coaching systems operating at roughly one million calls per month |
| ๐ฅ | Shipped a multi-tenant clinical LLM platform in under three months, including on-premise PII masking at approximately 97% accuracy |
- ๐งญ Flow-Next - An open-source delivery system for coding agents. Specs hold intent, fresh-context workers implement bounded tasks, cross-model reviews catch misses, and receipts show what happened. Keep a human in the loop or hand a reviewed spec to Ralph overnight. Claude Code ยท Codex ยท Cursor ยท Droid. flow-next.dev
- ๐งช flow-next-opencode - OpenCode port of Flow-Next with Ralph mode, re-anchoring, and multi-model review gates.
- ๐ง 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
- ๐๏ธ 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.
- ๐ DocIQ - Agentic legal document intelligence. Word-native tracked changes, research across six legal databases, reusable playbooks, data rooms, and zero-persistence court anonymization.
- ๐ sheets-cli - Composable Google Sheets CLI and agent skills for Claude Code and Codex.
- ๐๏ธ obsidian-skill - Agent skill for reading, searching, and maintaining Obsidian vaults through the official CLI.
- ๐ better-skill-builder - Production patterns for building portable Agent Skills.
- ๐ raindrop-skill - Bookmark search and management for Claude Code, Codex, Amp, and OpenCode.
- ๐ง outlookctl - Local automation for Classic Outlook when Microsoft Graph is unavailable.
- ๐ gmickel-bench - Real-world coding-agent evaluations with model judging and human review.
- โ๏ธ SmartTrim - macOS menu-bar utility that repairs mangled text copied from AI coding assistants.
- โ๏ธ claude-code-config - My working Claude Code configuration and reusable setup patterns.
- ๐ cursor-windsurf-convert - Lossless conversion between Cursor and Windsurf rule formats.
- ๐ค CodeWhisper - The 2024 predecessor to Flow-Next: end-to-end task implementation and fast codebase-to-LLM context assembly.
- ๐ง memorybot - Persistent-context Node.js AI agent, first released in 2023.
- ๐๏ธ turborepo-shadcn-nextjs - Bun and Biome-powered monorepo starter.
- ๐ฎ CodeQuest - Framework for building interactive learning games.
- ๐ฅ openehr-quest - OpenEHR learning game built on CodeQuest.
- ๐ก hugh - Node.js library for Philips Hue bridges and lights.
- ๐ค telegram-hue-bot - Telegram control for Philips Hue.
- ๐ CalSync - Archived macOS utility for copying events between Apple calendars.
- Agentic engineering: connect product discovery, implementation, review, QA, and measurement into one system that agents can run and humans can inspect
- 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
- Reusable infrastructure: turn recurring delivery problems into open-source tools other teams can run
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






