AI product
tm.post
An AI agent that runs social media for a small business
From idea to published post in one minute
- Role
- Product from zero: flow, architecture, AI workflow, launch
- Period
- 10 days
- Team
- Solo, with AI
- Status
- Shipped, scenarios being refined
How it works
01Event
The owner types, speaks, or sends a photo.
02Reading it
The system extracts the substance and applies the club's settings.
03Generation
Text and visual are produced together.
04Preview
The post is shown in full; edits happen in the same thread.
05Publishing
Once confirmed it goes to every connected channel at once.
What it is
tm.post is an AI agent that runs a business's social media instead of its owner. The first niche is tennis and padel clubs with two to four courts.
It works across Telegram, VK and MAX. There is no separate interface: the owner writes to a bot, and the rest happens without them.
Problem
Something happens at the club every day: a session, a tournament, a court freeing up, a shifted schedule, someone winning. There is no shortage of reasons to post.
Turning each one into a post is what nobody wants to do. Publishing slips, the channels go quiet, and the owner is back to being their own club's SMM manager.
How it is built
The owner tells the bot what happened — by text, by voice or with a photo. The system reads the message, writes the text, prepares the visual and shows the post in full. Edits happen in the same conversation, and nothing is published without confirmation.
Behind that conversation is not a single model call but a connected workflow: routing the message, generation, assembling the post, the content plan, the scheduler, delivery to the channels, reminders and error handling. The infrastructure is self-hosted and orchestrated with n8n; when a step fails the scenario retries instead of collapsing.
What the AI does
The model handles understanding and wording: “tournament tomorrow, two slots left” becomes a post in the club's voice, with the right call to action and formatting. It also produces the visual.
Everything involving time, money and delivery runs as ordinary code: the schedule, the queue, plan limits, posting to a specific platform. That split keeps behaviour predictable — the model can only be wrong where the mistake is visible before publishing.
How it works
01Event
The owner types, speaks, or sends a photo.
02Reading it
The system extracts the substance and applies the club's settings.
03Generation
Text and visual are produced together.
04Preview
The post is shown in full; edits happen in the same thread.
05Publishing
Once confirmed it goes to every connected channel at once.
06Rhythm
The content plan, the scheduler and reminders keep the cadence without the owner.
What already works
- Onboarding and channels
- Setup takes 30–60 minutes and happens once.
- Telegram, VK and MAX
- MAX comes with the relevant plans; one post reaches every connected channel.
- Text, voice and photo
- Three ways to say what happened — no phrasing, no headline required.
- Text and visual generation
- The post arrives assembled rather than as a draft.
- Preview, edits, confirmation
- Nothing is published without the owner's confirmation.
- Content plan
- A run of posts for a week or a month in one request.
- 10 personalisation settings
- Tone, address form, hashtags, emoji, image style and the rest are tuned per club.
- Fallback scenarios
- A failure at any step leads to a retry, not to a lost post.
Interface
Dashboard
Plan, publications left in the period, counters and the next post — the whole account state on one screen.
Generation settings
Topic, tone, address form, frequency, hashtags, emoji and image parameters. Set once, then it runs on its own.
Generated result
The post is shown in full — exactly as subscribers will see it — and goes out only after confirmation.
My role
Built the product from zero in 10 days — from the idea and the use case through the architecture, the AI workflow, the interface and a working launch.
Worked alone. I designed the product, the architecture, the flows and the logic; AI was the tool that made designing and writing it faster.
Stack
- Surface
- Telegram Bot
- Telegram Mini App
- Orchestration
- n8n
- self-hosted
- Channels
- Telegram
- VK
- MAX
Current status
The product is finished as a working system and has its first customers. Setup takes 30–60 minutes; after that the owner comes back to the bot only when there is something to say.
Existing scenarios are being refined. There are no figures on this page: none are verifiable yet, and unverified ones do not belong here.