Case study · Manufacturing

Troubleshot: from a one-week audit to a live, founder-owned platform in four months

When a machine goes down, the operator radios maintenance. That call holds the most useful context about the failure, and nobody writes it down. Troubleshot captures it automatically. Its founder had proven the idea on a lab setup. I started with a one-week audit, rebuilt it into a live platform, moved the radio capture to the cloud, and handed the whole thing over to his own account.

Book a free 30-minute call All case studies

1 weekAudit to a clear plan

2 daysRebuild to live

20 findingsFrom testing, all shipped

~4 monthsPrototype to hand-over

At a glance

A proven idea, turned into a product the founder owns

What started as a fixed-scope, one-week audit grew into a full rebuild, a hardware integration, and a hand-over to the founder’s own infrastructure. Every phase after the audit was driven by his own testing.

Client	Troubleshot LLC; founder Zach Kauffman
What it does	Captures plant-floor radio calls and turns them into context for the next technician
My role	Technical lead and sole engineer: audit, architecture, build, deployment, and ongoing support
Timeline	Late May 2026 (audit) to September 2026 (hand-over); support ongoing
Shipped	The web app (installs on phones), a docs site, a marketing site, and the upload software for TRN-1, the radio-recording device
Built on	Cloudflare, with OpenAI for transcription and summaries, and a Raspberry Pi radio receiver
Quality	114 automated tests, and releases that check themselves before and after going live

The starting point

A convincing demo, on one network

On May 22, 2026, the founder’s lab setup ran the full loop for the first time: radio call in, useful context out. It was held together by three separate pieces, all on one local network.

01 · The app

Lived in one browser

Every symptom, fault, and fix was saved in the browser itself. Nothing synced between people or devices.

02 · Transcription

Ran on a laptop

A small service on a laptop received the radio clips and sent them to OpenAI to be transcribed.

03 · TRN-1

A radio recorder

A Raspberry Pi with a radio receiver recorded the plant’s radio traffic and uploaded clips over the local network.

It couldn’t support a pilot customer. Zach hired me for a fixed-scope, one-week audit with three questions: what’s demo-ready, what will break in front of a customer, and what’s the fastest path to pilots without rebuilding for its own sake?

How it evolved

Keep what works. Rebuild everything around it.

The audit’s verdict was to keep the capture loop Zach had proven and rebuild everything around it on a real backend. The rebuild went live within two days. Every phase after that answered a specific need: a demo, a round of testing notes, an error in the device logs, or Zach taking full ownership.

Late May

Audit sprint. A one-week review of the prototype, its data, the AI workflow, and the demo setup, ending in a recommendation: keep the capture loop, rebuild on a real backend.

Jun 17–18

Platform rebuild, live on day one. A proper backend and database, a new web app, company sign-up, docs and marketing sites, and an in-app AI assistant.

Jul 6

Radio capture moves to the cloud. The laptop service was retired; transcription now runs inside the platform, and TRN-1 uploads to it directly.

Jul 7–8

Demo-readiness fixes. Log and escalate in one step, a findings workflow for responders, photos from the phone camera, and an Engineering tier on the escalation ladder.

Jul 14–16

Round two. All 14 findings from Zach’s testing shipped the next day. The app now installs on phones, and a review closed a permissions gap.

Sep 14

Device hardening. Traced the radio device’s occasional errors to their root cause and fixed it there.

Sep 28

Hand-over. Domain, database, files, and services moved to Zach’s own Cloudflare account, with a plain-language owner’s guide.

How I built it

So a small team can change it without fear

01 · One rulebook

Rules live in one place

The escalation ladder, shift handoffs, solution review, and who can see what are defined once. The server, the app, and the tests all use the same rules, so they can’t drift apart.

02 · No servers

Nothing to patch

Everything runs on Cloudflare. There are no servers to maintain, costs stay low at pilot scale, and the whole platform runs on a laptop for development.

03 · Many companies

Separate from day one

Each company gets its own workspace, tied to its email domain. Every record and file is kept separate by company, and audio and transcripts are visible to admins only.

04 · Root causes

Fix the cause, not the symptom

The radio device logged about 8 errors a week out of roughly 115,000 check-ins. I traced them to brief database hiccups on a step the device never needed, removed that step, and added targeted retries.

05 · Releases

Releases can’t skip steps

One command runs the tests, backs up the data, deploys, and checks the live site, and prints how to roll back if anything fails.

06 · Reviews

Regular review sweeps

Structured reviews looked for permission gaps and fragile code. One closed a gap on approved solutions, and the automated test suite grew to 114 tests.

Live · ready for pilots Troubleshot / 2026

A product Zach owns end to end, and can demo to customers

What used to need three programs on one local network is now one hosted platform.

An operator logs a fault, presses Escalate, and makes their usual radio call. When the technician opens the issue, a context card built from the call’s transcript is already waiting. No extra work from the operator.

Talk about your prototype

It started with a short, fixed-scope audit, the same first step as my AI-built app audit.

What Zach has now4

01Context captured automaticallyThe radio call becomes a transcript-backed card for the next technician

02A complete maintenance workflowSymptoms and faults, an Operator → Maintenance → Controls → Engineering escalation ladder, shift handoffs, and a reviewed library of step-by-step fixes with photos

03Live everywhere it needs to beThe web app on phones and desktops, a public docs site, and a marketing site with working contact and demo forms

04Full ownershipEverything runs in his own account. Every record and all 37 stored files were checked against the originals, and a plain-language owner’s guide sits alongside the code

The result, judged

One-week audit turned into a clear plan to reach pilotsPass
Rebuilt platform live within two daysPass
20 findings from two rounds of testing, each shipped within a day or twoPass
Handed over to the founder’s own account, every record verifiedPass
Pilot customersUp next

Your prototype

Have a demo that works, but not in front of a customer?

A one-week audit tells you what’s demo-ready, what will break in front of a customer, and the fastest path to pilots.

Book a free 30-minute call AI-built app audit