A whole online school inside one stack: sales funnels and leads, scheduling, finance and subscriptions, courses with homework, a messenger and live classes with recording. Permissions are not hard-coded — the school owner assembles roles, student contacts are masked under Russian personal-data law, and changes land in an audit trail. Every school runs on the shared platform: it gets its own subdomain, and data is separated per school on every request to the database, the search index and the queues.
One source of reference data for every system in the company. Structure and record versions are kept as immutable snapshots: any past state can be restored and served over the API. Access goes through corporate OIDC and LDAPS, and every change lands in an audit trail that cannot be rewritten.
FastAPI
PostgreSQL
OIDC
LDAPS
Kafka
Versioned EAV
Shared directories with versioning and change history.
One catalogue across the website, Telegram and MAX.
System diagram
Storefront & bots
Catalogue & orders
Inventory
Payment
Delivery
MOSSALUT
Client project
A Next.js storefront and two retail bots — in Telegram and MAX — on top of the Vendure commerce core. There is one catalogue: products, prices and stock live in the core rather than in two places at once — so a basket filled on the storefront or in chat can actually be paid for. Inventory is wired to the accounting system, and delivery is quoted through several carriers.
It collects organisations from external sources, verifies their sites and contact channels, removes duplicates and scores leads. Personal outreach is written by a model and sent over email, Telegram, VK and WhatsApp — with rate limits, blacklists and a sequence that stops on the first reply.
The layer where AI features live as part of the product instead of as a call to somebody else’s API. Long scenarios run durably and survive a restart, tools are only invoked through an access policy, the model is chosen by a gateway, and every call is visible in the log: what was asked, what it cost, what came back.
Temporal
MCP
LiteLLM
OPA
pgvector
OpenTelemetry
AI workflows with access policies and execution logs.
System diagram
01Interface & BFF
02Control plane
03Durable runtime
04Tools & policies
05Models & knowledge
06Logs & audit
AGENT PLATFORM
Client project
All our projects
Swipe through the rail. Open a project to see the brief, scope and live website.
01/16
Drag the rail or use the arrows
Not just interfaces.The systems working behind them.
Tools
Tools are chosen for the task, not the other way round.
Next.js
React
TypeScript
Node.js
PostgreSQL
Redis
Docker
Next.js
React
TypeScript
Node.js
PostgreSQL
Redis
Docker
1C
MoySklad
Ozon
Wildberries
Payments
Telegram Bot API
AI / LLM APIs
1C
MoySklad
Ozon
Wildberries
Payments
Telegram Bot API
AI / LLM APIs
What we can build for you
04 services
01020304
01
02
03
04
XYZ
01 / 04
Launch a website
From a clear storefront to an order: customer journeys, interfaces, catalogues, payments and delivery.
02 / 04
Bring sales into a CRM
Connect enquiries, tasks and teamwork in one product with roles, permissions and an activity history.
03 / 04
Automate workflows
Connect systems through APIs and events so your team can stop moving data by hand.
04 / 04
Add AI to your product
Integrate models into search, data processing and workflows with access control and execution logs.
01 / 04
01Launch a website
02Bring sales into a CRM
03Automate workflows
04Add AI to your product
01
Launch a website
From a clear storefront to an order: customer journeys, interfaces, catalogues, payments and delivery.
02
Bring sales into a CRM
Connect enquiries, tasks and teamwork in one product with roles, permissions and an activity history.
03
Automate workflows
Connect systems through APIs and events so your team can stop moving data by hand.
04
Add AI to your product
Integrate models into search, data processing and workflows with access control and execution logs.
on the work
Studio / AVT TEAM
How we agree on the work
A small team runs the project end to end: design, code, integrations and servers in one pair of hands. Here is how we agree on the work — before the first line of code.
4
directions: websites, CRM, processes, AI
1
team for design, code, integrations and servers
0
vendors assembled piece by piece
01 / 04
01AVT TEAM
What happens after an enquiry?
We start by understanding the task: existing systems and the expected outcome. Then we agree on scope, milestones and project terms.
Enquiry
3 fields: name, contact, task
We clarify
task, systems, outcome
We agree
scope, milestones, terms
02AVT TEAM
How is the price determined?
The price depends on scope, integrations and requirements. Deliverables, timelines and payment terms are recorded in the agreed proposal and contract.
It depends on
scope, integrations, requirements
We record
deliverables, timelines, payment
Where
proposal and contract
03AVT TEAM
How are code and design handed over?
Deliverables, access credentials and rights to the results are agreed in the contract before work starts. A product should not depend on whether we answer an email.
We hand over
code, design, documentation
Access
passes to the client
Rights
in the contract before the start
04AVT TEAM
What happens after launch, and with an existing product?
Support scope, changes and maintenance terms are discussed separately and recorded in the contract. An existing product can be improved: we start from it and your task.
