Lead capture through to cash collected, on autopilot. One CRM, one ledger, one dashboard, 89 workflows running live, and a nightly check that catches a bad number before anyone acts on it. Outside that: SMS engines, AI agents, web apps, and a stack of personal systems. If there is an API, I have connected it.
A trading education business running weekly live webinars into an application, a booked call, and a close. Phone setters, DM setters, closers. I own the entire technical and process layer: lead capture, the CRM, attribution, the money, the reporting, and everything in between.
Nobody presses a button for any of this. This is the clock the business runs on.
| Every 2 minutes | New registrants land in the CRM, deduped, tagged, and flagged net-new so a rep can call them while they are still warm. |
| Live, on event | Bookings, applications, confirmations, cancels, reschedules, disqualifications, and no-shows alert the team the second they happen. |
| 8:00am | Yesterday's reported numbers are checked against the real record. Missing forms, duplicates, and mismatches get posted with the responsible rep named. |
| 10am, 1pm, 4pm, 8pm | Per-rep dial counts pulled from the phone system and posted to the floor manager. |
| 9:15pm | Closer and setter leaderboards post themselves, with commission already calculated at each rate. |
| 11:45pm | The payments ledger reconciles against the CRM and the rep forms. Anything that does not tie out gets flagged before morning. |
| Session day, 7am | Team hype message goes out, and the registrant reminder sequence arms itself. |
| Session day, afternoon | Attendance swept from the webinar platform, every lead sorted into watched, left early, or never showed, and tagged in both the CRM and the email tool. |
| Sunday 10am | The replay goes to everyone who missed, on a list rebuilt from source data, using a link that updates itself after every event. |
| Month end | Commission block and the operations report roll up on their own. |
Every rep walks into a call with a full business audit of the prospect already written into the CRM. Crawled, analysed, and filed before anyone picks up the phone.
67 systems, grouped. Use the filter to jump to what you care about, and open Details on any card for the full build.
Anyone can connect two apps. These are the failures that do not throw an error, the ones that quietly corrupt a month of numbers before anybody notices. Each of these came out of production.
A leaderboard started returning nothing. The CRM's search API caps cursor pagination near 10,000 records and threw a 400 that a try and catch swallowed, so it failed quietly instead of loudly. Fixed by chunking the query per calendar day so every cursor loop starts fresh and never approaches the ceiling. Verified across 29 days, 80 requests, and 12,620 records with no failures.
The old automation created a brand new duplicate lead on every single opt-in, so the CRM count was inflated and no report could be trusted. Traced it to four specific steps, killed them, and made one workflow the only thing allowed to create a lead, with search-before-create so a lead missed on one pass self-heals on the next instead of vanishing.
A commission block showed one rep at zero despite obvious sales. The cause was a nested array formula that only ever evaluated the first element and returned zero for anyone whose name had two spellings in the data. Replaced with additive per-variant lookups. The lesson is that a wrong number with no error attached is far more dangerous than a crash.
Three different systems reported three different cash figures. I reproduced all three to the dollar, and the gap was two rows. One had a date cell containing text instead of a date, which meant every date-range lookup silently dropped it, because text is never greater than or equal to a date. Now caught by a validator before it reaches a report.
An automation returned money values as strings and it read like a data-type bug. It was not. That is rendering, not storage. The rule now is to verify against unformatted values before declaring a discrepancy, and to re-read after a write because reads lag. This one rule has prevented several chases after problems that did not exist.
A fully static site needed to create a real CRM lead, with nowhere to configure headers and no server to run. Solved by posting as plain text so the browser treats it as a simple request and skips the preflight entirely, with the allow-origin header set on the response. Zero infrastructure added.
A domain went live but the SSL certificate hung in a pending state. The cause was ordering: the domain had been attached to the host before its DNS pointed there, so verification cached a failure. Removing and re-adding the domain issued a fresh certificate in seconds. Attach after DNS resolves, never before.
A logging step was writing an API response into a column meant for a user's message, because it sat downstream of another node and inherited that node's output. Reference the originating node explicitly rather than whatever happens to be current when anything sits downstream of a branch. Silent column pollution is invisible until someone reads the report.
Audited live against both production instances on 29 July 2026. Grouped by what they do rather than what they are called. Open a group or search across all of them.
Counts audited live against both production instances on 29 July 2026. Test harnesses, scratch workflows and superseded duplicates are excluded from the 106. A green dot is running right now; a grey dot is built and either deliberately paused, seasonal, or superseded.
Not a list of things I have read the docs for. Every one of these is authenticated and running in something that is live.
The reason these systems hold up is not that nothing breaks. It is that nothing breaks twice. On my own systems audit, documentation scored the highest of ten categories.
One per platform, written from production rather than from the docs. What the pagination actually caps at, which endpoints are undocumented, which operations the API will silently refuse, where the cache lies to you. These are the notes that make the second build fast.
One per shipped system, each with what was built, what broke, why, and the rule that came out of it. When a client asks why a number looks a certain way six weeks later, the answer is written down rather than reconstructed.
A master operations hub linking every form, sheet, dashboard, channel and automation. An SOP library. Campaign libraries. An onboarding system. A full systems audit with a weighted grade card that scored my own build honestly and named what would fail at three times the volume.
106 workflows. 89 running live. 50 plus APIs in production. I take a problem, connect the right tools, and ship something that runs without you touching it.