Telematic 2.0: Why We Rebuilt Our Own Product from Scratch

September 28, 2026

Last year we told you about Telematic.Pro, our service for automating Telegram channels. A lot has changed since then: in early 2026 we made a tough call and rewrote the product from scratch. Today it’s no longer an auto-poster — it’s an AI newsroom that writes posts, assembles videos and publishes them across ten platforms on its own. Here’s why it happened and what it taught us.

Telematic.Pro 2.0

When a product outgrows its architecture

The first version was built in three months and handled its original job perfectly well: collect material from sources, rewrite it with AI and publish it to Telegram. Users came — and brought new requests with them:

“Can it also post to VK and YouTube?”
“Can it turn an article into a video?”
“How do I know what’s working best?”

Every one of these requests ran into the foundations. Version one was built around Telegram, with 17 tables, a single AI provider, two languages and not one automated test. Bolting video and new platforms on top of that would have been like adding a third floor to a foundation poured for a shed.

Version one and version two

Rewriting doesn’t mean starting over

We kept what mattered most: our understanding of users, the workflows that had already proved their value, and the domain itself. Everything under the hood changed:

an API on NestJS and Prisma, a Next.js front end and our own design system;
63 data models instead of 17 — platforms, video, scripts, payments, analytics;
background processes moved into dedicated workers, so heavy jobs never slow down the interface;
multiple AI providers instead of one — each task gets the model that suits it best;
7 interface languages and 164 automated test suites from the very first months.

The first five weeks went into the core: sources, AI processing, the scheduler, platform integrations and payments. From there, the product grew on solid ground.

Choosing a content source

Video: the hardest part — and the most valuable

The biggest new feature in version two is producing video from text. AI writes a script and breaks it into scenes, the text is voiced with a synthetic voice or your own, captions, music and animated inserts are added, and then the video is edited together and published to YouTube, Shorts and social media.

Behind it are decisions users never see, but which make the product economically viable:

distributed rendering — videos are assembled on separate machines that connect to the platform and stand in for one another if something fails;
local speech recognition with Whisper instead of a paid API — a noticeable cut in the cost of every video;
AI cost tracking for every operation — so we can see exactly what each post and each video costs the product.

Video studio

Video render

Autopilot: a product that works without the user

The best-case scenario for a client is to set everything up once and forget about it. Autopilot picks up fresh material from sources, prepares posts and videos, and sends them for approval or straight to publishing. According to the counter on the service’s website, autopilot has already published over 8,600 posts.

Autopilot

Growth as part of the product

Another lesson: marketing works best when it’s built into the product, not bolted on beside it. Version two added a free post generator that needs no sign-up, pages for specific use cases and competitor comparisons, a knowledge base in 7 languages — and, for the team, a Growth hub with lead discovery, email A/B testing and AI-powered reply analysis. Today the service has around 1,000 pages in the search index.

Free post generator

When a rewrite from scratch is justified

A rewrite is an expensive decision, and more often than not the right move is to evolve the code you already have. But there are signs that the time has come:

1.
New features break old ones. Every improvement demands workarounds in the foundations.
2.
The data model no longer describes the business. The product already does more than its database can express.
3.
There are no tests, and touching the code is scary. Development slows down with every release.
4.
You know exactly what to build. Version one has already tested the hypotheses — version two is built on knowledge, not guesswork.

Telematic ticked all four boxes. Over eight months and more than 1,400 iterations, the product grew from a Telegram auto-posting tool into a full-fledged AI newsroom.

The detailed case study with screens is in our portfolio, and you can try the service itself at telematic.pro.

If your product is also running up against its own architecture, tell us about your project and we’ll help you decide what pays off better: evolving it or rewriting it.