AI Agent
DentalBot
CRM for a solo dentist inside Telegram: appointment planner, patient base, and an agent that books a patient from a voice message — it checks the calendar and writes the record itself.
NestJS · Next.js 15 · Tool calling · TMA SDK
Back to services
[AI AGENTS]
AI agents, built end-to-end.
The agent takes the routine work inside your system: books clients, triages tickets, prepares reports. It runs under your rules — permission boundaries, a spend cap, and a log of every action.
Tool calling · NestJS + Next.js
Discuss Project // FEATURES
What's Included.
Acts, not advises
The agent calls your tools: calendar, CRM, database, email. The result is a record in your system, not a message in a chat.
Permission boundaries
What the agent may and may not do is stated explicitly. Irreversible actions require a human confirmation.
Spend cap
A limit per conversation and per month. The model provider's invoice will not surprise you at month end.
Every step is visible
The log shows what the agent called, what came back, and what the decision was based on. There is something to review when it gets things wrong.
Backend and interface from one person
I write the NestJS endpoint and the Next.js screen myself. No second contractor to hire and coordinate.
Your data
Three options: the provider's API, a model on your own server, or stripping identifiers before sending. We pick what your requirements allow.
// SCREENSHOTS
What it looks like.
// EXAMPLES
Work Examples.
Relevant projects
AI Agent
Medical assistant for clinic patients
Patients arrive prepared: before the visit an LLM assistant asks about symptoms, explains how to get ready, and assembles a report for the doctor. Appointments are pulled from the clinic's medical system.
React · TypeScript · Python · YClients API
AI Chat
Aida — AI Support
AI customer support assistant interface: order card right in the chat, quick actions, typing indicator, topic sidebar.
Chat UI · UX · AI
// PROCESS
How I Work.
Transparent process from first message to launch
01
Requirements
Discuss the task, goals and requirements. Form a clear technical specification.
02
Analysis
Research the domain, choose architecture and tools.
03
Development
Write code, show intermediate results. Iterative process.
04
Launch
Deploy, test, hand over the project and documentation.
// FAQ
FAQ.
Answers to common questions
A chatbot replies with text; an agent does the work. In DentalBot the agent books a patient from a voice message: it hears the name and time, checks the doctor's calendar, and writes the appointment into the database. It has 29 tools for that — calendar, patient base, notifications.
The price depends on how many tools the agent connects to and on how expensive its mistakes are: an agent moving calendar entries needs fewer checks than one that sends money. I quote after reviewing the task; the initial estimate is free.
Dangerous actions sit behind a confirmation: the agent prepares the operation, a human approves it. Everything else is in the log, so you can see where it went off and fix the rules. Agents are never error-free, so the question is always the cost of a mistake and who catches it.
With a review: we look at which tasks an agent genuinely removes and which are cheaper to solve with ordinary code. Often half the list needs no LLM at all, and it is fairer to say so upfront. You get a work plan and a stage-by-stage quote.
Claude, GPT, Gemini, and open models hosted on your own server. The choice follows the task, the token price, and your data requirements. I build the architecture so the model can be swapped without rewriting the system.
// RELATED SERVICES
Often ordered together.
Telegram Mini App
Turnkey mini-app development. Shops, services, dashboards, games — any functionality with Bot API.
Learn moreMVP for Startups
Rapid product launch from idea to working prototype. 5+ successful MVPs.
Learn moreMaintenance & Support
Bug fixes, small tasks, improving existing projects, urgent help. Billed hourly, quick start without lengthy negotiations.
Learn more