▲
AI systems and LLM assistants
Production RAG, local LLMs that keep data inside your perimeter, embedding into CRM, messenger bots, voice interfaces.
Qdrant vLLM Mistral Qwen
Custom development since 2026
We build
complex
software
to order.
The things off-the-shelf SaaS does not cover: AI assistants on local LLMs, custom B2B platforms, replacements for foreign ERP systems, marketplace aggregators. Production development with responsibility for the outcome — not prototypes, and not wrappers around a chatbot.
5+
production projects launched in fintech, e-commerce and legaltech
4 weeks
to a working prototype, at the earliest
1–5M ₽
typical turnkey project budget — we quote in roubles
Stack
Python · TypeScript · React · Postgres
Integrations
Accounting · Payments · CRM · Marketplaces
AI
Qdrant · vLLM · Mistral · Qwen · hosted APIs
Infrastructure
On-premise · your cloud · your servers
What we do
Six areas where we are strongest.
We take on work that sits where real difficulty meets real business value. If the task is standard, we will point you at a product or an agency that fits it better.
◆
Custom B2B platforms
Customer portals, CRM, ERP, WMS, marketplace aggregators. Complex workflows, role models, integrations.
FastAPI Postgres React WebSocket
◉
Replacing SAP and Oracle
Migration from foreign ERP and CRM to local alternatives. Accounting integration, on-premise deployment, regulatory readiness.
On-premise Phased migration
▼
Marketplace integrations
One dashboard for several marketplaces. Two-way stock synchronisation, automated price management, reconciliation.
Marketplace APIs ClickHouse
■
SaaS platforms and multi-tenant
Subscription services with billing, recurring payments, pricing plans and multi-tenant architecture.
Billing Multi-tenant White-label
✶
Custom analytics and BI
Management dashboards, real-time metrics, demand forecasting, unit economics by SKU, channel or cohort.
ClickHouse Grafana Metabase
Cases
What these projects look like in numbers.
B2B / Manufacturing
Scenario
−92%
time spent searching a 12,000-SKU catalogue
AI assistant for a sales team: a quote in 12 minutes instead of 2 hours
RAGLocal LLMCRMTelegram
Legaltech
B2B2C
automation of a legal workflow
Partner portal for a bankruptcy procedure
FastAPIDOCXPaymentsReal-time
E-commerce
Scenario
+12 pp
margin, through automated repricing
Marketplace manager for a cosmetics brand
Marketplace APIsClickHouse
Cards marked Scenario describe how we would build the system and what it would change, based on our experience with similar work — not a delivered project. Cards without the mark are delivered work. Full case studies exist in Russian only for now.
Process
How a project runs.
No months of specification meetings. Two weeks to an architecture, and a working prototype four to eight weeks after that.
- 01
Discovery
One to two weeks. We dig into the problem, talk to your team, walk through the processes. We come out with an architecture and an estimate.
- 02
Prototype / MVP
Four to eight weeks. A working prototype covering the most important scenario, with a demo every two weeks.
- 03
Production
Eight to sixteen weeks. Tests, load, security, documentation, and training for your team.
- 04
Ongoing work
A monthly retainer. Support, new features, changes driven by feedback. We do not walk away after the release.
Why us
What separates us from two hundred other studios.
01
We do what SaaS does not cover
We are not competing with off-the-shelf CRMs or site builders. We go where standard products hit their limits and a thin wrapper around a chatbot does not work.
02
Production, not a prototype
We ship code with tests, migrations, monitoring and documentation. You get a system, not a demo that works on one laptop.
03
Numbers up front
Budget, timeline and risks at the start, not settled along the way. Scope changes go through short change requests.
04
We stay after the release
Most of a product's value appears six to twelve months after launch. We stay on a monthly retainer to build it.
FAQ
Common questions
What does MZR Digital do?
MZR Digital is a custom development studio. We build B2B platforms and SaaS products, AI assistants on local LLMs and RAG over corporate documents, replacements for SAP and Oracle, integrations with accounting systems, marketplace analytics and legaltech. We work remotely, from the first call to production.
What does an AI assistant over our own documents cost?
A RAG pilot on a limited document corpus takes three to six weeks. A production system with access roles, answer auditing and automatic index updates takes two to four months. GPU infrastructure is a separate line: monthly if rented, a one-off capital cost if bought. We quote in roubles, so ask us for a figure at the current rate.
Can SAP or Oracle be replaced without stopping the business?
Yes, if you move process by process through a parallel track: the new module runs alongside the existing system, both work together for a while, and only after reconciliation on real data does the process switch over completely. A pilot of one key process takes six to ten weeks, a phased migration six to twelve months. In a bank or a critical-infrastructure operator the timeline is different — three to five years, because of regulatory reporting and acceptance testing.
What is your minimum budget?
We take projects from roughly one million roubles upwards. Smaller tasks — tweaks, or extending an existing product — are usually unprofitable for both sides, and a freelancer will serve you better.
How long does a project take?
Two to six months for the full cycle from discovery to production. An MVP can be assembled in four to eight weeks if the scope is clear from the start.
Do you work on the client's own infrastructure?
Yes, especially for projects with sensitive data or regulatory constraints. We deploy on-premise, in a cloud of your choice, or on your own servers.
Can you join our team as outstaff?
We can, but that is a different model. We do product development and take responsibility for the outcome. If you need hands on a time-and-materials basis, we will point you to teams that fit better.
Which LLMs do you work with?
Hosted models and local ones from the Mistral, Qwen and Saiga families — chosen for the task and the compliance constraints. For compliance-sensitive projects, local only.
What about warranty and changes?
Six months of free bug fixes after release. Scope changes go through separate change requests with a fixed price and deadline.
Next step
Tell us about the problem.
We will say what it takes
and what it costs.
A call with the founder of the studio. No slide decks — we go through your problem and give you an estimate of the effort, the risks and the architectural options.