AI task tracker in Telegram via the Business API
A bot on the Telegram Business API watches client chats while AI spots tasks in the message stream and sorts them by project. Time tracking, monthly summaries and a ready-made brief for handoff to development.
The problem
Client support tasks come in through Telegram — direct messages and group chats, mixed in with ordinary conversation. In that stream tasks get lost, there is no logging of hours spent or monthly summaries per client, and to hand a task over to development you have to retell the context by hand. What was needed was a tracker that lives right inside Telegram and doesn't force the client to learn a new tool.
Before
- Tasks existed only as messages in the conversation — no statuses, no time logging.
- Hours per task were counted roughly; monthly summaries were assembled by hand.
- Handing a task off to work meant retelling the context and hunting for the right messages and screenshots.
What we built
- We connected a bot to the working account via the Telegram Business API — it sees messages in a chosen chat folder and can reply on behalf of the employee (always with an honest note that an AI assistant, not a human, is writing).
- AI (Claude) classifies incoming client messages: is this a task or not, and which project it belongs to. The bot suggests — a human always confirms with a button; tasks are never created automatically from a client's message.
- The bot learns from corrections: if the operator fixes the project or rejects a suggestion, later decisions get more accurate (few-shot on the accumulated corrections).
- Multimodality: the bot understands not only text but also the screenshots and PDFs a client sends — and extracts the essence of the task from them.
- Time logging rounded to 15 minutes, an automatic monthly summary by client and project, and one-command generation of a ready-made task brief for handoff to development.
Results
- Tasks stopped getting lost in the conversation — each one is captured with a status, hours and its original context.
- The client installs nothing and changes no habits: they still just write in Telegram, and all the mechanics live on the bot's side.
- Handoff to development is a ready-made brief with screenshots and history instead of a manual retelling.
- The AI cost is negligible: at typical load the model's budget fits within a few dollars a month.
The client's name is withheld under NDA. Real names and details are shared at the brief.
A similar task?
Describe it briefly — within 1–2 days we'll come back with an analysis and an hourly estimate. If a packaged solution fits, we'll say so honestly.