AI strategy that survives contact with production.
For companies that want AI in the product, not in the pitch: what to build, what to buy, what to fine-tune and what it will cost per user. Dmitrii has run machine learning in production since 2015 and LLM systems since 2019.
Four decisionsmade with you, in writing
Most AI budgets are lost on one of these four. Each gets a written answer with numbers in the first two weeks.
Build, buy or fine-tune
An API model, open weights or your own: chosen by the task, the data you hold and the cost of a wrong answer.
The price of a call
Cost per user and per call before launch; routing between models, caching and batching. Alion records the cost of every call.
A bar before release
A test set and a quality bar before the first release, and the same check before every release after it.
Data and security
What leaves your perimeter, what a model may see, and how prompt injection is stopped before it reaches a tool.
Three productswhere AI paid its way
From the work history on Dmitrii's Alion profile; the full cases are on their own page.
LLMs over business data
Services that read contracts, listings, support threads and market feeds and turn them into decisions people act on.
Signals for investors
AI that reads market data and produces investor signals, including a fintech platform for a UAE bank.
15 billion posts a day
Collectors and a machine learning layer that pull genuine casting and booking calls out of social networks in four languages.
Asked beforethe first call
What is a fractional Chief AI Officer?
A senior leader who owns AI decisions part-time: which models, what data, what it costs, how quality is measured and who builds it. It suits companies that need those decisions made well but not a full-time executive.
Do we need our own model?
Usually not at first. Most products start on an API model with good retrieval and evaluation; fine-tuning comes when the data and the volume justify it, and the plan says when that is.
How do you keep LLM costs under control?
By measuring the cost of every call from day one, routing easy requests to smaller models, caching what repeats and setting a budget per user before launch.
How soon is there a result?
A written plan in two weeks and one change in production within the first month.
Tell Dmitrii what you are building.
One message is enough: the product, the stage and what is stuck. He answers himself, usually within a day, and says plainly whether he is the right person for it.
An AI agent can do the same: check_fit with the offer, then send_message at alion.io/user/dm/mcp.
