Ch. 01 / 04AI Agents
Agents that do the work, not the demo.
Wiring an LLM to a prompt takes an afternoon. Making it hold up on the thousandth run — when a tool call fails, the data is messy and the edge case nobody wrote down finally shows up — is the actual job. I co-developed Sorch, a recruiting agent that does the screening calls recruiters used to make by hand, and I'm building Friday, an agent that runs real office work.
- multi-step, tool-using agents
- voice agents in production
- approvals, audit logs & guardrails
Ch. 02 / 04Workflows & Automation
Operations that run themselves.
Every business has work that moves between inboxes, spreadsheets and tired people. I map how it actually flows, then rebuild it as systems that run on their own. With Salex, a salon's whole operation — bookings, staff, stock, money — moved into one app for the owner and one WhatsApp chat for the customer.
- n8n, OpenClaw & Hermes
- WhatsApp-native flows in production
- connectors into Tally & office systems
Ch. 03 / 04Hardware / Physical AI
Intelligence with a body.
The real world doesn't fit in a context window. With Ahum Labs I'm building Project AI Box — a single-board computer that runs a team of sandboxed agents on local models, on a custom control board and firmware validated in simulation before it touches silicon. Friday is getting a body too: a 3-ft wheeled assistant robot.
- custom control boards & firmware
- IoT & embedded systems
- simulate first, then build
Ch. 04 / 04Full-stack & Infrastructure
None of it matters until someone uses it.
I ship on every surface it takes — web, Android in Flutter, desktop in Tauri — and I run my own infrastructure. Two retired PCs became a private cloud that's online 24/7 — I can deploy to it from anywhere, even from my phone.
- Next.js, Node, Flutter, Tauri
- Linux, Docker, AWS & DevOps
- a 24/7 homelab I run from my phone
