Principal Product Architect / Technical CPO — applied-AI systems & developer platforms
I build platforms that replace teams. 20+ years turning hard technical domains — agentic AI, payments infrastructure, developer tooling — into production systems, usually shipping solo or with a small senior team, then scaling them. My focus now: making LLM agents reliable enough to run in production, and the developer platforms around them.
🌐 cmdop.com · djangocfg.com · linkedin.com/in/cmdop · 📧 markolofsen@gmail.com
Letting agents act on live systems is easy to demo and dangerous in production — a plausible-looking action can break things. The hard part isn't the LLM; it's making agent actions trustworthy at scale. That problem is what I build around.
📝 Read the deep-dive: How I made AI agents safe to run on real infrastructure — the eval/instrumentation loop behind Cmdop.
Cmdop — AI agent & automation platform
An agent → gRPC → server → SDK platform with a multi-agent runtime: agent-to-agent handoff, tool-calling governed by structured-output contracts, and human-in-the-loop control. Every tool call runs through an eval/instrumentation loop — the contract is validated before execution, the full trace is logged, runs are scored on tool-call validity / task success / unintended side-effects, and brittle steps get guardrails + automatic retry/failover. That loop is what turns "a demo that works once" into a runtime you can trust to widen agent autonomy. Drives thousands of concurrent agent sessions over persistent gRPC/WebSocket streams. Node.js / Python / React SDKs.
DjangoCFG — AI-native Django framework
Replaces Django's settings model with a type-safe Pydantic v2 system and unifies Django + Next.js admin + payments + realtime into one architecture. It's AI-native: capabilities are exposed through an MCP server, so coding agents query what the framework can do and its schemas directly instead of scraping docs. Compresses first-deploy SaaS setup from weeks to ~30 seconds.
A crypto exchange engine (sub-50ms order matching, multi-asset wallets/ledgers) · an AI real-estate ROI platform for a Top-3 Canadian firm (~50% faster assessments, ~$150K/yr saved, ~40% more qualified leads) · an AI Cloud IDE (VS Code in browser, AI agents, MCP) · an Unreal Engine 5 Pixel Streaming framework.
| Project | What it is |
|---|---|
| openrouter-commit ⭐ | AI commit-message generator, 300+ LLM models |
| django-cfg ⭐ | Type-safe Django config (Pydantic v2) |
| litellm2 | LLM orchestration wrapper |
| carapis-parsers | High-performance automotive data parsers, 20+ platforms |
| metaeditor | React toolkit for Unreal Engine 5 Pixel Streaming |
| docusaurus-doctor | Diagnostic CLI for Docusaurus |
15+ open-source libraries total — API client generators, automation CLIs, dev tooling.
Applied AI — LLM orchestration · agents / multi-agent runtimes · MCP · RAG · structured outputs · LLM evals & observability · HITL Platform & distributed systems — gRPC · WebSockets · high-load real-time · durable execution · idempotency/consistency · SDK & API design Full-stack — Python · Django · FastAPI · Pydantic v2 · Next.js · React · TypeScript · Node.js · Go · PostgreSQL · Redis · ClickHouse Cloud — AWS · GCP · Azure · Docker · Kubernetes · Cloudflare FinTech / other — exchange & trading systems · ledgers · Ethereum/TRON/DeFi · Unreal Engine 5 / WebRTC
Currently: building applied-AI platforms (Cmdop, DjangoCFG) — agentic systems made reliable enough for production. Open to: Principal Product Architect / Head of Applied AI / Technical CPO / VP Engineering roles, and fractional / advisory engagements. Remote (APAC, UTC+8; strong EMEA overlap), open to relocation.