Distributed systems that don't fall apart — and AI that actually ships.
Self-hosted AI sprints and infrastructure audits for SMEs. GDPR-ready, no cloud lock-in, results in 2 weeks.
Three ways I can help.
AI Sprint
Your company running on private, self-hosted AI in 2 weeks. OpenWebUI + Ollama, 2 n8n automations wired to your tools, team training, full handoff.
- GDPR-compliant, no cloud lock-in
- 2 automations connected to your ERP/CRM/Slack
- Team onboarding + maintenance playbook
Architecture Audit
Find what breaks before it breaks. 2-hour deep dive into your stack: reliability, cost, security, single points of failure.
- Prioritized fix list with effort / impact
- Real postmortems from similar systems
- Actionable within one sprint
Fractional SRE
Keep your systems alive without a full-time hire. Architecture, cost control, incident response, and ongoing reliability.
- Weekly check-ins + async monitoring
- Incident response within business hours
- Monthly reliability report
Projects I've built
from scratch.
Full-stack DePIN platform for crowdsourced network intelligence. Users scan WiFi, cellular, BLE and IoT networks via iOS and earn tokens on Solana proportional to data quality and geographic scarcity.
Read case studySelf-evolving AI companion that runs entirely on-device. Phi-4-mini (3.8B, 4-bit quantized) on Apple M1 with just 8GB RAM. No cloud, no subscription. Uses MCTS reasoning, self-play evolution, curiosity-driven learning, and overnight sleep consolidation.
Read case studyHonest static analysis for Python — catch circular imports, god modules, and tight coupling before they become postmortems. Open-source core on PyPI plus a paid Pro Kit.
Project page →12-point Infrastructure Health Check
The 3 things that cause 3 AM outages — and how to spot them before your users do. PDF checklist + self-scan questions.
Try a private LLM in your browser
On-device LLM running via WebGPU. Zero API calls, full privacy — the same stack I ship in the AI Sprint.
Interactive experiments.
Proof-of-concepts, tools, and on-device demos. Everything runs directly in your browser.
Latest writing.
Every Improvement I Made to My 1.7B Agent Made It Worse
Four attempts to make a small tool-calling agent smarter, three of which made it measurably dumber — and the design rule that fell out of the numbers.
Read →I Wrote a Self-Healing Script That Could Strand the Machines It Healed
Renumbering a remote site with no console. Seven machines recovered, three went dark — and the failure I blamed on hardware was two bugs in my own safety net.
Read →A 2012 Mac Mini Beats a Xeon Per Core
Twelve workers and 76 threads built from nine second-hand Mac minis and a used PowerEdge — then the benchmark inverted how I place jobs.
Read →Let's build
something together.
Book a free 30-minute intro call. We'll map your AI or infrastructure problem and pick the right engagement.