// michael dominick's engineering company

The systems an LLM can't write and a cloud vendor won't run.

The Mad Botter is accountable for the parts of a production stack that are hard to hire for — undocumented legacy systems, hardware and the edge, regulated and air-gapped deployment, self-hosted infrastructure we operate rather than hand off. Not a prototype. The thing that has to be running on Monday.

Writing code got cheap. Operating production systems didn't. Anyone can generate something that works in a weekend now; almost nobody can take a twenty-year-old schema nobody can explain, a packing line that talks over serial, and an export-control boundary, and be accountable for the result. The scarce thing moved from code to accountability.

See the work Talk to us

$ stty -F /dev/ttyUSB0 9600 raw && xxd /dev/ttyUSB0

02 4c 4f 54 3a 34 34 37 31 1d 03

x undocumented framing · no vendor spec

 

$ psql legacy -c '\d job_hdr'

jh_no | jh_cst | jh_st | jh_dt7 | jh_flg2

x 41 columns · 3 documented · author left

 

+ this is the job

// illustrative — not customer data

100% US-based engineering · A decade of production delivery · Direct US Navy and Air National Guard work

// we run what we build

// the work

What we've actually shipped

Specifics rather than capabilities, because you can check specifics. All of this is in production.

// aerospace / mro

Corrosion alerts on parts nobody was watching

Field gateways — BeagleBone Black and Raspberry Pi over ethernet, Wi-Fi, or cellular — authenticating with rotating BCrypt-hashed keys and streaming temperature, humidity, pressure, vibration, and moisture into a time-series store. A rule engine evaluates per-asset thresholds including N-consecutive-breach conditions, each with its own cooldown, and grades alerts to email and Slack. High-value aerospace inventory had been sitting in storage with no monitoring at all.

// aerospace / mro

MIL-STD-129 labels straight to the warehouse printers

Compliance labels used to mean opening a third-party tool, screenshotting the output, and eyeballing the print scale. Now they render as print-ready PDFs and stream to Zebra printers as ZPL with Code 128 and Data Matrix/PDF417 barcodes. RFID tag records cross-reference the job, the label, the shipment Transportation Control Number, and the DoD scan.

// aerospace / mro

14,000 line items out of fifteen Monday boards

An order physically migrated from board to board as it aged — RFQ, orders, warehouse, finance, archive — stitched together by hand-typed reference numbers, alongside a 1.6 TB document store with no relationship to anything. Migrated into one relational job that keeps a single identity from quote to delivery.

// heavy civil

A 300-ton crawler crane and a portable generator, same data model

Every asset across five operating companies got an identity and a weatherproof QR tag. Inspection cadences are enforced per equipment type — cranes on a different schedule and checklist than vibratory hammers — and transfers between subsidiaries run through request-and-approve with timestamped logging. Producing a certification history for an OSHA inspector went from hours of digging through filing cabinets to a scan.

// agriculture

Software that talks to the packing line

A Python/Flask platform with 500+ endpoints and 70+ models across a multi-database PostgreSQL architecture, integrated with AWETA sorting and grading lines, automated bin fillers, scales, and barcode scanners. Windows desktop and Android apps on the warehouse floor, at loading docks, and at employee kiosks. Self-hosted over WireGuard with GitLab CI/CD — we run it, not just ship it.

// agriculture

Recall as a query instead of a fire drill

Lot traceability through the entire chain: harvest bin to filler to pallet to shipment to customer. Cold-chain verification captured at close, and the cleaning, waste, and pest-control records certifications demand generated as a byproduct of the work rather than a binder someone fills in later.

Written up in full: the aerospace deployment and the heavy civil fleet system.

// open source

Or skip us and read the code

The work above is private, so you'd have to take the write-up on faith. These you can just read.

// ruby dredger-iot

Running in production in both our aerospace and heavy civil deployments — you can read the hardware layer itself. GPIO and I2C abstractions for embedded Linux, pluggable sensor drivers, retry and scheduling helpers, and config-driven pin mapping for BeagleBone. Real backends via libgpiod and Linux i2c-dev, plus a simulation backend so it runs on a laptop. Published as a gem, MIT.

github → // postgres · duckdb PG-Lake-Make

Turns PostgreSQL into a lakehouse: query Parquet on S3-compatible storage directly through Postgres using DuckDB's engine, backed by MinIO or SeaweedFS. Runs on a laptop for development or your own VM for production — which is the entire point.

github → // ruby Alice

An open data pipeline platform — connectors for PostgreSQL, Snowflake, DuckDB, and flat files, a visual query builder alongside raw SQL, and scheduling and monitoring for the transformations in between.

github → // c csv_to_parquet

A small CLI that converts CSV to Parquet through DuckDB, with type inference and correct handling of quoted fields, embedded newlines, and delimiters inside quotes. Unglamorous and useful.

github →

// what we take on

The parts that are hard to staff

// 01 Systems nobody documented

Twenty-year-old schemas, undocumented business rules, the database one retirement away from orphaned. An LLM cannot help you here — the knowledge was never written down, so it was never in the training data. Someone has to sit with it.

// 02 Hardware and the edge

Sensor gateways, label printers, barcode scanners, scales, packing-line equipment, kiosks in coolers. Code that has to be right because the machine does something physical when it runs.

// 03 Regulated and air-gapped environments

ITAR and EAR boundaries, CMMC and NIST posture, facilities where "just put it in the cloud" is not an available answer. We are designed to support your compliance posture; we do not certify compliance.

// 04 Running it, not just building it

Deployment, observability, upgrades, the pager. Most shops hand you a repo and leave. The part that actually matters is the two years after go-live, and that is the part we take.

// how we work

Bounded ownership, not bounded hours.

A retained relationship where we own a defined part of your stack — not an open-ended hourly arrangement, and not a one-shot project handed over at the end. You know what we're responsible for, we're the same people month over month, and the thing we're measured on is whether it runs.

We're small and senior by design. If you need twelve contractors next quarter we're the wrong call, and we'll tell you so on the first conversation.

x staff augmentation · hours with no ceiling x a repo handed over and a wave goodbye + a defined part of production, owned

// who you'd be working with

You probably already know how I think

Michael Dominick at his desk

I'm Michael Dominick. I've hosted Coder Radio for over a decade — 150,000+ developers a month — and I write at dominickm.com. If you've listened, you already have a decade of evidence about how I think about software, which is more than you'll get from any pitch.

The Mad Botter is the company I run. We're a small, senior, US-based team that builds and operates systems in places where the software touches something physical and the consequences are real — aerospace and MRO, heavy civil, industrial agriculture.

coder.show → dominickm.com → github →
// coder radio · latest episode 654: Prof Andrew Seely // live from the feed · 51 min · jul 18, 2026
▶
coder.show →
// dominickm.com · jul 24, 2026 AI SDRs Are Automating the Wrong Half of Sales // dominickm.com · jul 15, 2026 The $100,000 Cron Job // dominickm.com · jul 1, 2026 The 4th of July Compute Hack: The Cynical (But Brilliant) Economics of Anthropic’s Fable 5 Launch // live from dominickm.com/feed

// the ask

If you have a system like this, I'd like to hear about it.

Specifically: something in production that's load-bearing and badly understood, hardware you need talked to, a deployment that can't go to the cloud, or a system whose original author left. That's the work we're good at and the work we want.

If that's you, say so and we'll talk. If it isn't, the subscribe box below is the honest ask — most people reading this don't need us today.

Talk to us

// faq

Straight answers

What does The Mad Botter actually do?

We take ownership of the parts of a production stack that are hard to hire for: undocumented legacy systems, hardware and edge integration, regulated and air-gapped deployment, and self-hosted infrastructure we operate rather than hand off. Michael Dominick's engineering company, ten-plus years of production delivery, 100% US-based.

How do you engage — projects or retainer?

A retained relationship where we own a defined part of your stack, rather than an open-ended hourly arrangement or a one-shot fixed-price project. Bounded ownership, not bounded hours. Senior people, and the same people over time.

What does it cost?

It depends entirely on what we are taking ownership of, so we scope it against your actual systems rather than publishing a number that would be wrong for almost everyone. Working that out is not billed.

Isn't this the kind of thing an LLM can do now?

Some of it, genuinely. Writing code got cheap. What did not get cheap is being accountable for a system in production — an undocumented schema nobody can explain, a packing line that talks over serial, an export-control boundary, and a deployment that has to survive Monday morning. That knowledge was never written down, so it was never in a training set.

Who have you built for?

Three projects directly for the US Navy and one for the US Air National Guard, plus subcontract work supporting the US Army and US Air Force. On the commercial side: a DC-area aerospace parts broker and MRO shop, a Cincinnati-area deep foundations contractor running five subsidiaries, and a large industrial agriculture operation whose entire software stack we built and continue to operate. The first two commercial engagements are written up in detail; the third is under wraps beyond what is on the agriculture page.

Where is the team?

100% US-based engineering, and the same people you meet are the people who do the work. Michael hosts Coder Radio, so you can hear how he thinks about this before you ever talk to him.

// contact

Tell us what you're trying to build.