nknaveed.io
open a line
§ Lead Developer · Kannegiesser ETECH

Naveed Kakal.

Naveed Kakal, in a navy knit sweater, looking at the camera
fig. 00 — on the day shiftQA dept. not pictured

From the sensor to the boardroom.

For twenty years I've written the software that runs industrial laundries: the biggest ones, washing for hospitals, hotels and airlines around the clock. I've been lucky to work on every layer of it: the layout on the drawing board, the sensors on the floor, the controller programs that run the machines, the screens operators work from, the labor tracking that measures their work fairly, the plant's own reports, a hand in the cloud rollup that ties 400+ plants together, and the app that buzzes a supervisor's phone. That range comes from being on a very small team with projects on several continents. You get to understand the whole plant end to end, and you know the customers first-hand, on their floors, not through a ticket queue. I'm grateful for that. It taught me how every piece fits together, and why. Electrical engineer. IC at heart. Lately, the nights belong to AI.

day shift peaks at noon

Lead Developer, Kannegiesser ETECH. Material handling.

  • Twenty years writing software for the hardest stuff to move: dirty laundry.
  • Every layer, from a sensor on a conveyor to a push notification in a pocket.
  • Hands-on. I'm an IC at heart.
night shift peaks at 11 p.m.

20+ new products since May, mostly after bedtime.

  • An AI maintenance system running in real laundries.
  • Two paid games on the App Store. A science channel my daughters pick topics for.
  • A personal software factory that builds while I sleep.
based
Naperville, IL
on the floor
since 2006
trained as
electrical engineer
on the record
10,134 commits, one product line
QA department
Zoya & Maya
§ 01 — the job
twenty years of 3 a.m. calls, in three switches

Material handling on hard mode.

Cartons have corners. Laundry doesn't, and it changes weight as it's washed and dried. Now move it through a building the size of a big-box store, around the clock, with one system running every move.

If you run a warehouse, a factory or an airport, this is your stack too: the same sensors, controllers, routing and reports, with different nouns. Try to break the line.

fig. 01 — a laundry line, simplified tap a ◆ sensor to force it on (a test tool), again to let go
Laundry in, dirty and heavy. Watch the tags: heavier when wet, lighter when dry, then folded.0 through
an illustration, not a real plant. Tap a stopped item to ask it why.
§ 02 — the stack, bottom up
eight layers · one person · since 2006

Every layer of a plant. I've written code in all of them.

Most software careers are one cubicle on one floor of the building. I've worked every desk on every floor, and I take the stairs: from the bit a sensor flips to the report on the boardroom table.

L0 the plan · layout & roadmap

Before the first sensor: the drawing, and the long plan.

I'm in the room while a plant is still a layout drawing. I work with our sales team so the layout really meets the customer's needs, and I bring the gotchas they can't be expected to know: what our controls can and can't do, and which choice on paper turns into a headache in year three.

I also try to keep the software from becoming a rule book. It grows with the pressure each new plant puts on it, and I run its long-term roadmap, so even small, fast changes land inside one bigger plan: giving laundries what they need to run. We don't just program these plants. We're part of how they get sold, specified and built.

layout designworking with salescustom requirementsproduct roadmap
L1 I/O · field devices

Everything starts as a bit.

Out on the floor are the field devices. A photo-eye sees an item go by. A proximity switch confirms a gate is closed. A scale weighs a load, and in this business weight is money. A motor drive moves a conveyor, a reader tells one kind of item from another, and a switch sends each down its own track. Each of those is a wire, an I/O point, and network modules gather them onto the plant network. One plant I support maps 8,700 I/O points into a single controller, about 7,700 of them simple on/off signals.

In early 2009 I started the habit that became our testing standard: click to force, a test tool for commissioning a plant, not something left on in production. Click any sensor in the software and the plant acts as if a real item were sitting there, so whatever is coming up behind it stops and waits. Paired with a test PLC, it lets us exercise the whole stack, from the controller up to the reports, before a single real item moves. Tap a ◆ on the conveyor to force it on; tap again to let go.

photo-eyesproximity switchesscalesmotor drivesnetworked I/O
L2 machine control

I don't hand-write the PLC programs. I generate them.

The PLC is the plant's industrial controller: it reads the sensors and runs the motors, many times a second. When I started, we wrote those programs by hand, plant by plant. One of the first things I did was turn our standards into a generator: describe a plant once and the program writes itself. Not just a parts list of conveyors, lifts and panels: how each one is configured, and the network of what feeds what, and for what purpose. The detailed interactions between devices are modeled, so a wide range of capability comes standard instead of being written fresh each time. Fewer errors, much faster projects, and a small team that could take on far more of them at once. Today it covers 129 device types, with a hash check that proves a change didn't move a byte it shouldn't.

Talking to the PLC has had three lives: SOAP from Rails in 2007, an Elixir service I wrote in 2019, an OPC UA bridge today. One rule survived all three. Never act on a stale sensor. Read it again, live, before releasing anything.

B&R controllersgenerated programsOPC UAModbus TCP
L3 screens & workflows

Then a person has to run it.

In 2007 the operator screens were a per-seat license. The floor wanted eyes everywhere; the invoice said pick two. So we made the screen a web page. Nearly twenty years on, that page is eVue, and it's still what operators work in every shift at 400+ plants: a live picture of the whole plant, every item moving through it, and alarms that say in plain words why something isn't moving. In warehouse terms it sits where the control and execution systems (WCS, WES) would; in a factory, where the MES would.

One system runs every move across the whole building: the plant behind that I/O count has 611 rail sections and 93 machines, every one routed by eVue.

No two laundries work the same way, either. A hospital plant, a hotel plant and an airline plant each sort, store and send work through in their own order. So eVue is configured, not rebuilt, for each one: every station shows its operator just what they need to do that task right, not one generic screen for everything.

eVuelive plant viewconfigured per plantsince 2007
L4 people & productivity

Then there's the people.

Machines report on themselves. People are harder, and labor tracking is a different product with a different job: knowing who worked which station, on what, for how long, and measuring it in a way the floor accepts as fair. It takes understanding how a plant actually runs its shifts, not just its machines.

Operators log in to a task at a station. Their work is measured the way it really happens: per piece where that's fair, as a team in pounds per operator-hour where it isn't, and some jobs, like catching at the end of a line, are scored by the pace of the people feeding them. Breaks and shifts are built in, logins don't churn every time a customer changes, and supervisors get it by text and email, not just on a screen.

operator loginspiece rate & team ratebreaks & shiftsproductivity reports
L5 local reporting

The plant reads its own numbers.

Every time an item passes a sensor, a machine finishes a load or an operator logs in, it's a row in a database. So the plant can answer for itself: output by the hour, how long each machine runs, how busy each station is, and where the day is going wrong while there's still time to fix it. Plus a 6 a.m. report email someone actually opens.

All of it lives on the plant's own server. When the internet goes down, the floor doesn't notice.

KPI dashboardslabor standardsscheduled reports
L6 cloud reporting

Then the company reads all of them.

Each plant runs on its own. A hub at every site rolls its numbers up to the cloud every minute, so the people running a company of plants see one picture instead of forty phone calls.

The fleet is the hard part: 400+ plants, each its own building, floor plan and set of machines. This layer is a team build: Jason and Larry carried most of it, and I've added my share.

edge hub per plant1-minute rollupmulti-tenant warehouse
L7 mobile

And finally, a phone in a pocket.

Plant networks are walled off, and they should be. Pulse is how the plant's data gets out anyway, securely: a supervisor links a phone by scanning a QR code at the plant, then gets the numbers that matter, and a push when something is really wrong.

It's the top of the stack, and the part people actually carry. That's the whole building, bottom to top.

PulsePWAweb pushQR-linked devices
fig. 02 — one plant, eight layersL0 · the plan
§ 03 — the long run
twenty years, one domain

I stay for the results.

The decisions I make this year are the ones I get to check in year three. That's the part I like.

A lot of software gets put in place by someone who's gone before anyone knows whether it worked. I'd rather stay and take care of it. My main codebase started in 2007 and has never been rewritten, so every lesson from year one shows up in a later version, and every call I make today is one I'll be around to validate, or to fix.

Before laundry there was a car wash: nine years at Royal Oaks in Shoreview, Minnesota, ending as general manager. In 2006 an industrial controls company hired me, fresh out of electrical engineering, on a hunch that car washes and laundries have a lot in common. They do. Machines, water, valuable things moving through a tunnel on a schedule, and nobody can stop the line.

I left once, for two and a half years of Chicago tech at Centro. I wrote better software there and solved fewer problems per developer dollar, got promoted to engineering manager, and learned I'm an IC at heart. Then a prank phone call brought me home, just as Kannegiesser acquired ETECH.

Today I'm Lead Developer at Kannegiesser ETECH. The building can't stop, so the software can't either, and I'm still the one who gets on the plane.

20 yrs
one domain: industrial laundry
10,134
commits on one product line · Mar 2009 → Sep 2026
2007
the year my main codebase started. Never rewritten
400+
plants on eVue, each its own building and machines
four flights — most of my best stories start with a boarding pass
LAX2009 · seat: whatever's left

Commissioning a brand-new platform in Los Angeles.

We slept in our cars the first night. It was that close to not making it.

slept in the car
SAP2010 · San Pedro Sula

Building the largest sock factory in the world.

At the time, the biggest sock manufacturing plant anywhere. Different product, same problem: soft goods that all have to end up in the right place.

biggest in the world
UKCOVID · cabin: nearly empty

A 747 with almost nobody on it.

Flew to the UK to help start up a laundry for the health service while the rest of the world stayed home.

started it up
HKGOct 2026 · 36-hour window

Upgrading the laundry behind an airline's hub.

Hardware, software and network, plant-wide, all at once, in a 36-hour window. Most of it was running in under 20. Read the case file →

ahead of schedule
field rules — how I design, learned the hard way
R-01 · learned at Centro

Count problems solved per developer dollar.

Well-run teams can still produce a flotilla of blah. The ledger that matters is what the floor stopped suffering.

R-02

Remove the seams.

Responsibility leaks out through the joints between specialists. I'd rather own the stack top to bottom than hand off at every floor.

R-03

Speak the floor's language.

Use the words operators use, not the ones engineers invent. When the software talks like the floor, the floor can tell when it's wrong.

R-04

Let reality win.

When the plant and its digital picture disagree, ask a human what happened, and offer only answers that can physically occur.

R-05

Chase the bottleneck, not the fashion.

I moved my analytics service from Rails to Rust in a day. I kept the PLC bridge in Node: the PLC is the slow part, and a rewrite would reopen nine solved production bugs.

R-06

Improve the condemned.

Push a dying system as far as it will go. Its real limits, not the rumored ones, become the spec for whatever replaces it.

shift change — when my code goes in, by hour

I like work.

The day job peaks at noon. My own builds start once the QA dept. (Zoya & Maya) is asleep, and peak at 11 p.m.

day job my own builds
§ 04 — the night shift
Oct 2025 → Oct 2026

Twenty years of ideas. Now I build them.

AI coding finally let me build ideas I've carried through a long software career, and tinker with things I never thought I'd have the time to learn. It started two years ago, copy-pasting from ChatGPT; this year it became a fleet. The AI I care about most is still for the people on the floor.

Weave · AI maintenance for laundries · in production

A tech scans the machine. The machine tells them how to fix it.

A maintenance tech is responsible for dozens of machine types, and half get serviced less than once a year. Nobody remembers the safe procedure for an annual deep clean. In Weave, they scan a QR code on the machine and get it, step by step, written from the manufacturer's manual with the page cited, in plain English (Spanish when asked). The crew already talks in a group text, so Weave joined the group text. The owner gets a plain-English report every Monday.

1,832 commits since Dec 19, 2025 · built for people who don't want more software · open case file →

One person who knows the floor. A fleet of agents that knows the code. I decide what ships.

Since July, most of it runs through Burrow, a software factory I wrote from scratch: agents and autonomous workflows that run in production on my own machines, 1,436 runs across 29 feeds so far. Work comes in from tickets, cards and timers, walks a path I drew, and comes back done, or with a question only I can answer. A model from a different vendor reviews every change it writes. When it isn't sure, it parks. It never guesses.

That's how a bit more effort turns an idea into something that runs every week: a Monday industry digest, a nightly product review, a science channel. Every run is costed.

The agents write the code. I'm still the one who gets on the plane.

fig. 03 — Burrow, condensed from its own map: 29 feeds, 18 graphs, 27 profiles. Current shows where work moves. Pink is the authority that decides, and it's never the model that wrote the change. Swipe the map.
§ 05 — the binder
night shift ’26

Everything I built after bedtime.

Graded on what holds up: tests, users, receipts. Not how fast it shipped.

six to start with, then the range · swipe the six for more

Also in the pile: Roll-for-it dice for family decisions · sites for a few more people I like (Glow Garden, M&F Laundry) · and the interview skill that writes § 06.

§ 07 — open a line
back on the floor at six

Got a plant, a product, or a problem?

Operators, vendors, peers. I'm easy to reach.

— Naveed. Built with a fleet of agents, mostly after bedtime.
QA dept. (Zoya & Maya) not pictured.

nk@naveed.io — zsh