AvtoUM AI command center
Project year2026
AvtoUM
AI SAAS · PRODUCT ARCHITECTURE · FULL STACK

Your website attracts visitors.AvtoUM turns them into customers.

A digital employee that uses company knowledge, helps the visitor make a decision, captures a contact and hands the manager a prepared context.

THE PROBLEM

The invisible gap before a request

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.

ECONOMIC PRINCIPLE

Work better with traffic already paid for

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.

PRODUCT JOURNEY

From page view to prepared request

01

Visitor

Opens a product or service page.

02

Context

Permitted signals help identify the question or hesitation.

03

Dialogue

The AI employee consults using company knowledge and rules.

04

Lead

With consent, it captures contact details and creates a structured request.

05

Handoff

A manager receives the dialogue and can take over without losing context.

NOT A CHAT WIDGET

The dialogue is connected to business operations

Ordinary AI widget
  • Waits for a question
  • Uses one general instruction
  • Ends with an answer
  • Leaves the dialogue outside operations
AvtoUM
  • Uses permitted context signals
  • Combines platform rules with company knowledge
  • Creates a contact and structured request
  • Connects CRM, Pulse, billing and human takeover
ENGINEERING EVIDENCE

Four systems that make the product difficult

One messaging pipeline

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

ONE CONNECTED PRODUCT SYSTEM

Pulse: operational visibility

Visitors, dialogues, leads, pages and topics turn the website from a black box into an observable work surface.

ONE CONNECTED PRODUCT SYSTEM

Human takeover

When AI reaches its boundary, a manager joins the same conversation and the product prevents conflicting replies.

ONE CONNECTED PRODUCT SYSTEM

AI cost control

Model calls are separated by task, company and cost. Limits and graceful degradation protect both the platform and client data.

ONE CONNECTED PRODUCT SYSTEM
SYSTEM ARCHITECTURE

Not one prompt — a full SaaS contour

Frontend

React 18 · Vite · Tailwind

Backend

Python 3.12 · FastAPI · SQLAlchemy

Data

PostgreSQL · Redis

Realtime

WebSocket · heartbeat · presence

Background work

APScheduler · reports · classification

AI

central model policy · prompt layers

Payments

ЮKassa · subscriptions · limits

Infrastructure

Docker Compose · nginx · SSL · RU VPS

BUILT END-TO-END

From the first customer message to SaaS economics

AvtoUM combines customer communication, operational work and platform economics. These are not disconnected screens: they form one business process and one production architecture.

Customer communication
  • Proactive website widget
  • Business knowledge and multi-layer prompts
  • Need qualification and structured lead capture
  • Human takeover in the same dialogue
Business operations
  • Contacts, requests, tasks and notes
  • Cross-channel dialogue history
  • Realtime Pulse and topic analytics
  • Manager and administrator workspaces
SaaS economics
  • Plans, trial periods and dialogue limits
  • YooKassa payments and subscription lifecycle
  • AI cost accounting by company and operation
  • Graceful AI pause while CRM data stays available
MY ROLE

Founder, product architect and full-stack developer

I formed the product concept and designed the frontend, backend, CRM, realtime contour, AI pipeline, prompt model, billing, AI economics, administration and production infrastructure.

Key development contours

  • Production LLM engineering beyond prompting
  • Realtime, WebSocket, billing and infrastructure
  • Product, economic and operational thinking
  • The ability to unite architecture and full-cycle delivery

Have an idea for a digital employee or another complex AI challenge?

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