Technical founder and full-stack architect. Two companies built from scratch — one that scaled, one that didn't survive COVID. Currently running a 45+ service homelab as an AI-native operations proving ground, building multi-agent orchestration tooling, and writing about the things I've learned (mostly the hard way).
| Metric | Value |
|---|---|
| Companies founded | 2 |
| Funding raised | $2M (Pangeam, at $15M valuation) |
| Cloud cost reduction | $240k/year (Statricks, cloud to bare-metal migration) |
| Infrastructure scale | 45+ Docker services, 20+ TB storage, 40+ PostgreSQL databases |
| Current lab cost | ~$150-200/month |
Statricks — Co-founder, Technical Lead
Built a classified ads analytics platform from zero to production on bare-metal infrastructure.
- Designed and operated multi-TB PostgreSQL HA clusters on bare-metal servers
- Built ensemble ML pipeline for classified ad categorization and pricing
- Led cloud-to-bare-metal migration that cut infrastructure costs by $240k/year
- Maintained zero unplanned downtime for 18 consecutive months
- Full ownership: architecture, deployment, monitoring, on-call
Pangeam — Co-founder, Technical Lead
Built computer vision and ML systems for enterprise facility management. Raised $2M at a $15M valuation.
- Deployed on-prem CV/ML systems in 2 Fortune 50 headquarters
- Integrated 50+ LIDAR sensors for real-time occupancy tracking
- Navigated enterprise procurement cycles and security reviews
- COVID killed the in-office occupancy market. We returned ~$400k to investors and shut down cleanly
- The technology worked. The timing didn't.
Running a homelab as an AI-native ops environment — not as a toy, but as a real proving ground for how AI agents can operate infrastructure.
- 45+ Docker Swarm services on consumer hardware (RTX GPUs for local ML inference)
- Claude Code as operator — using MCP protocol to give AI agents direct infrastructure access
- ZFS snapshots as an AI safety mechanism (rollback when agents break things)
- Building initiative-engine — a multi-agent orchestration platform that turns conversations into parallel workstreams
Stack
Languages: Python, TypeScript, SQL, Bash
Frontend: React, Next.js, Tailwind CSS
Backend: FastAPI, Hono, Node.js
Data: PostgreSQL, Redis, Elasticsearch, Milvus (vector DB), Neo4j
Infrastructure: Docker Swarm, ZFS, Traefik, Nginx, Prometheus, Grafana
AI/ML: Claude Code, MCP protocol, LiteLLM, local inference (Ollama)
Tools: uv, pnpm, Git, SonarQube, Ansible
Blog coming at blog.johnefreier.com — retrospectives on startup failures, homelab architecture decisions, and building with AI agents.
- initiative-engine: Multi-agent orchestration platform with 10+ specialized AI agents
- Homelab AI ops: Giving Claude Code direct infrastructure access via MCP servers
- Writing: Honest retrospectives on what worked, what didn't, and why
jefreier@berkeley.edu · LinkedIn · X
