Express review
10,000 ₽
A 45-minute call and a written summary: which model fits, where to run it and what it will cost per month.
If you order the selection within 30 days, the review price is credited toward it.
Choosing infrastructure for LLMs
Your own LLM for your business
I help you figure out where to run a large language model (LLM) for your company: on your own computer or server, in a Russian cloud or through a ready-made API. If needed, I set it up end to end.
- Remote
- Individuals and companies
- Contract and receipt
- NDA
Discuss on Telegram
The bot asks 6 short questions about your task, and I reply within a business day.
Who it is for
- You want "your own ChatGPT" for employees but are not ready to hand documents over to a third-party service.
- Law firms, clinics, accounting, manufacturing: contracts, medical and financial records, technical documentation.
- You have already bought a graphics card or a server and do not know which model it can handle.
Three ways to run an LLM
01
API in Russia
A ready-made service: no hardware of your own, cheap when requests are few and the data can leave your premises.
02
GPU rental
A cloud in Russia: your own copy of the model on a rented server, paid monthly.
03
Your own hardware
Your server: full control over the data, pays off under heavy load.
If you do not need your own hardware, I will say so.
Models I deploy
Models
- Qwen
- DeepSeek
- Gemma
- Llama
- gpt-oss
- Mistral
- GLM
- Kimi
- GigaChat
- YandexGPT
Tools
- Ollama
- vLLM
- LM Studio
- Open WebUI
- OpenAI-compatible API
GigaChat and YandexGPT: their open versions.
Formats and prices
Selection with a cost estimate
from 20,000 ₽
2–3 options (API, GPU rental, your own hardware), configurations, provider prices, cost over 12 months, risks under 152-FZ, the Russian personal data law.
Testing models on your documents: plus 10,000 ₽.
Model installation
from 25,000 ₽
The chosen model on your computer or server and an Open WebUI chat for employees, without document search or integrations.
Turnkey launch
from 70,000 ₽
The model on your server or in a Russian cloud, a chat for employees with document upload, an API for your systems, a guide and a week of support.
CRM integrations and advanced search over a document base are estimated separately.
Ongoing support
from 10,000 ₽ per month
Up to 2 hours a month, updates and answers to questions within a business day.
You pay for hardware and cloud rental directly to the provider. No VAT is charged.
How we work
- 01You press the button and answer 6 questions from the bot.
- 02I reply within a business day and suggest a format.
- 03Payment, a call, a written summary.
About me
- MLOps engineer: I build platforms for running neural networks.
- Since 2022 I have been building GPU clouds and inference platforms.
- Talk "Choosing infrastructure for LLMs" at the Ya pro backend conference (Yandex).
- Speaker at HighLoad++, DevOops and PyCon. I write articles on Habr.
Questions
Will our data go out to the internet?
Not if the model runs on your side or in a Russian cloud. That is the kind of setup we pick.
Will our graphics card be enough?
I will tell you after the review: it depends on the model, the number of users and the length of the documents.
Why pay when there are free guides?
Installing the software is easy. The hard part is choosing a model and hardware that can handle your task without overpaying several times over.
Let us talk about your task
Discuss on Telegram
The bot asks 6 short questions about your task, and I reply within a business day.