One messaging pipeline
The channel receives and sends messages; knowledge, qualification, CRM and billing logic stay in one engine.


A digital employee that uses company knowledge, helps the visitor make a decision, captures a contact and hands the manager a prepared context.
A website sees visits but not hesitation. A manager learns only about people who decide to write or call. Product comparison, repeated price checks and uncertainty remain outside the sales process.
AvtoUM works with traffic the business already pays for: it helps visitors make a decision, qualifies demand and turns the conversation into an actionable request for the team.
Opens a product or service page.
Permitted signals help identify the question or hesitation.
The AI employee consults using company knowledge and rules.
With consent, it captures contact details and creates a structured request.
A manager receives the dialogue and can take over without losing context.
The channel receives and sends messages; knowledge, qualification, CRM and billing logic stay in one engine.
Visitors, dialogues, leads, pages and topics turn the website from a black box into an observable work surface.
When AI reaches its boundary, a manager joins the same conversation and the product prevents conflicting replies.
Model calls are separated by task, company and cost. Limits and graceful degradation protect both the platform and client data.
React 18 · Vite · Tailwind
Python 3.12 · FastAPI · SQLAlchemy
PostgreSQL · Redis
WebSocket · heartbeat · presence
APScheduler · reports · classification
central model policy · prompt layers
ЮKassa · subscriptions · limits
Docker Compose · nginx · SSL · RU VPS
AvtoUM combines customer communication, operational work and platform economics. These are not disconnected screens: they form one business process and one production architecture.
I formed the product concept and designed the frontend, backend, CRM, realtime contour, AI pipeline, prompt model, billing, AI economics, administration and production infrastructure.
I build multi-task AI agents, continuously develop their skills and integrate them into real business processes — from the first hypothesis to production.
Send the task