I'm an operator at Methodology, a food manufacturing company in San Francisco. I didn't come up through engineering — I came up through the work. Kitchens, floors, delivery routes, labor plans, closes. I learned software and AI architecture because the systems I needed didn't exist, and waiting for someone else to build them wasn't working. So I built them myself.
The day job, systematized
01
Weekly floor OS. The operating rhythm for a production week: nine lanes, five physical closes, two labor ledgers. What used to live in people's heads now lives in a system.
02
Internal ops platform. FastAPI + Next.js + Dagster over Supabase. Letters review, RAG over company knowledge, labor planning, deliveries. One platform instead of a dozen spreadsheets.
03
Data engineering. ETL pipelines and DuckDB transforms, so the numbers the floor runs on are the same numbers the business runs on.
04
Delivery operations. Tooling to audit a full week of delivery instructions across 860 stops — a job that used to be impossible to do by hand.
05
Medically Tailored Meals. Research and execution toward Medicare Advantage MTM. Food is medicine; manufacturing it well at scale is an operations problem worth solving.
The self-learning line
An AI operations layer that ingests company knowledge, trains on it, dispatches agents, checks their work against ground truth, and corrects itself. It runs like the line it replaced — so that's how I drew it:
The fleet
charlieONLINE
gpu workhorse
RTX PRO 6000 96GB Threadripper · 128GB
nerveONLINE
orchestrator
M4 Mini · 10W runs 24/7
muscleONLINE
storage + fallback
RTX 5090 · 16TB
omarchyDRIVING
my hands
laptop · Claude Code where I drive it all
zero inbound ports · whitelisted commands only · git history is the audit log
How I work
Domain first. I build for problems I've personally had. The spec is the scar tissue.
Self-taught, on purpose. Engineering, ML, agent architecture — learned by building, at night, against real deadlines.
Use what you build. I dogfood everything. If I won't run my week on it, it isn't done.
Auditable or it didn't happen. If a system acts, there's a record. No magic, no black boxes.
Small fleet, real work. You don't need a data center. You need the right loops running all the time.
Where this goes
Software agents coordinate the operation today. The same control plane — the command bus, the evaluation loops, the constitution — is what robotic cells on the floor will answer to tomorrow. That's the long game: the best-run food manufacturing operation in the world, built by the person who worked the floor, not someone who visited it once.
drew@omarchy~ $fleet status charlie .... ONLINE · nerve .... ONLINE · muscle ... ONLINE site ...... served from the edge · loop .... running weekly