I replaced my own software vendor, then kept the platform

My real estate company ran on Podio and GlobiFlow, and so did my business partner's brokerage. In 2025 Podio repriced. Podio cut automation actions from 5 million to 25 thousand a month and started metering the API. The bill went from $25/month to about $250/month, for software that was already only doing most of what we needed.

I decided to replace it instead of paying the increase. What came out of that is the platform both businesses run on today — and it is the same thing I now build, smaller and faster, for local businesses that have the same problem in a different shape.

The second install is the one worth paying attention to. Writing software for yourself proves you can build. Standing one up for a partner's business, where somebody else's payroll depends on the follow-up going out, is a different job — and it is the one that looks like what I would be doing for you.

• 65,575leads under management
• 2businesses running on it
• 1,434commits since July
• 35operator runbooks

These are not numbers from a pitch deck. This is a system I built and still run — one instance for my own company, one for my business partner's brokerage — and it has been sending real texts and real email every working day since July.See how it works.

How it is put together

• DataPostgreSQL · NocoDB
• OrchestrationTemporal · TypeScript worker
• CloudAWS Lambda · SQS · S3 · CloudWatch
• Voice and messagingTwilio · ElevenLabs
• EmailMailgun
• Data enrichmentEndato skip tracing · Google Street View
• Mail and alertsLob · Telegram
• InfrastructureCloudflare Access · Cloudflare Tunnel · Docker · Proxmox

How it behaves in production

The interesting part of an automation platform is not that it sends things. It is what happens when something goes wrong at two in the morning.

It will not send by accident

Sending is gated three ways, and all three gates are edited by a human in a grid rather than by a deploy. A bad value never opens the gate wider — it falls back to a tighter default. The unattended daily scan sends only when a specific field reads exactly ARMED; anything else is a dry run that reports what it would have done.

It survives the machine dying mid-send

Workflows are durable. A reboot in the middle of a campaign resumes where it left off rather than starting over or silently dropping the rest. That is the single biggest difference between this and a chain of Zapier steps.

Nothing is lost while the building is offline

Third-party events buffer in AWS and are pulled down when the server is reachable again. An unrecognised event is captured rather than dropped, so a new provider can be added later and its backlog replayed.

How it stays on the right side of the law

This is the part most owners are quietly worried about, and the part the cheap tools do not talk about.

• Do Not Call registry. Every number found by skip tracing is checked against the registry before it is ever used, and anything flagged is opted out for texts and calls alike.
• Opt out means opt out. A reply is matched against a configurable list; one match opts that number out everywhere it appears, not just on the lead that replied.
• Only mobile numbers get texted. Every number is classified, and a landline is never texted because it happened to be listed first.
• Send windows and rate limits. Outbound goes out during business hours at roughly one message a minute, not in a burst that reads as spam to a carrier.

Why there are no screenshots on this page

Every grid in that system has real people's names, addresses and phone numbers in it. That is not mine to publish, and a vendor who would publish it is telling you exactly how they will treat your customer list.

A walkthrough built on a demo dataset is in progress. Until then, I will show you the live system on a screen share during the call, which is better anyway — you can ask it questions.