Ilia Ilinskii · analytics & AI consulting

I get your data in order — and put AI to work on it

An analyst who builds the AI systems too

7+ years in analytics at Samsung, Okko, Recharge and Nexters ($GDEV). I set up analytics and attribution, find growth points, and build AI systems on top of your data: knowledge bases, assistants and analytics agents.

I work personally — you talk to the person doing the work, with no account managers in between. For bigger jobs I bring in a small team of specialists I trust.

Ilia Ilinskii

First call is free · fixed monthly budget — if the work takes longer, that’s my risk · remote, CET · English and Russian · USD or EUR

What I do

The three jobs people bring me most often. Each starts with an audit, which on its own gives you a clear picture and a plan.

Analytics

Analytics you can trust

Numbers in reports don’t match, and nobody can say which marketing actually pays off. I put it in order and find where the money is.

  • Audit: which events and numbers you can base decisions on, and which you can’t
  • A warehouse and dashboards that agree with billing
  • Attribution that shows which channel really pays back
  • Growth points in the funnel and the payment flow

AppsFlyer · Firebase · GTM · AppMetrica · ClickHouse · Superset

Analytics × AI

AI on your own data

The team wants AI, but the models know nothing about your company. I give them reliable context — and they start doing real work.

  • A knowledge engine: agents collect material, check facts and keep the base fresh, so models answer from your knowledge instead of making things up
  • Ask your data: a manager asks an AI assistant and gets numbers from the warehouse, with sources — no ticket to an analyst
  • Assistants and agents in Telegram and your work tools that take routine off people

multi-agent ingestion · fact checking · RAG · MCP · ~13,000 pages in production

AI

LLM product audit

An AI feature is already live, but it answers unreliably or costs too much. I find out why and fix it.

  • Review of architecture, context and cost
  • Hunting down the bugs and cases where the assistant breaks
  • A test set for prompts, and iterations based on the results

evals · prompt architecture · cost · debugging

Analytics experience

Seven years inside product and marketing teams: mobile apps and games, streaming, EdTech and payments — from startups to a NASDAQ-listed company.

Career

Nexters ($GDEV, NASDAQ)2024 — now
Product owner and tech lead, attribution. Hired and led a 4-person ML team: MMM, retention prediction, and a measurement framework whose results informed marketing budgets. Now advise the team.
Recharge.com, Amsterdam2022 — 2024
LTV models and a forecasting pipeline; channel ROI presented to senior management; SEO modelling behind 10% growth in organic customers.
Okko2020 — 2022
Analyst to team lead of 6 within a year. Attribution and cohort frameworks behind +10% acquisition year over year and −10% CAC; delivery time halved.
Samsung Electronics2019 — 2020
Price-elasticity and trade-in models; data collection automated across 100+ partners.

Client projects

AI EdTech · since 2026 · ongoing

From a tracking audit to an AI assistant

We started by auditing the tracking. Today the team has one analytics view across web and app, a knowledge base its agents keep up to date, an AI assistant in Telegram, and AI access to the data — managers ask the assistant for numbers instead of asking an analyst.

Mobile entertainment apps · 3 projects · 2025 — 2026

Analytics and attribution from zero

Built analytics from scratch and set up mobile attribution (AppsFlyer, Firebase, GTM, AppMetrica), then looked for growth points. In one project I found a problem in the payment flow worth up to +15% ARPU.

+15%ARPU potential found in a payment flow

AI experience

Whatever I build for clients, I first build and test on my own products at oshn-ai, my AI product studio.

~13,000
pages in knowledge bases running on my engine
5
live knowledge bases for 3 organisations
4
AI products of my own, from idea to release
20
languages Close Reader speaks

Knowledge engine

Used by clients

A Rust engine with multi-agent ingestion: it pulls in Telegram, sites, YouTube, drives, audio and images, checks facts and keeps a wiki where every claim has a source. Models reach it over MCP, behind OAuth.

Close Reader

Live

An AI reader for iOS and Android: explains a word in context, reminds you who a character is, answers about the book — never looking past your page.

Visit closereader.app

Const.

Live

A personal AI agent across WhatsApp, Telegram, iMessage, Slack and email. Payments, new contacts and deletions wait for your OK.

Visit const-agent.com

Oshn Meditation

Relaunching

An iOS AI coach in 10 languages: an LLM writes each session around the user’s state, ElevenLabs voices it, and billing runs in the same pipeline.

How I work

A fixed monthly budget: you pay for the approved plan, and if the work takes me longer, that’s my risk.

  1. Free intro call

    One hour on your task and the scope, free of charge. Afterwards I tell you what the audit will take, in hours and money.

  2. Audit

    I look at what you have today: data, tracking, AI systems, processes.

  3. Plan

    Scope, timeline and hours per month. You approve it.

  4. Work

    Billed monthly by the plan. Each month closes with an act of completed works.

  5. Anything extra

    Work outside the plan is estimated separately and starts once you approve it.

Serbian sole proprietorInvoices in USD or EURMonthly act of completed works

The first hour is free

Let’s talk

Tell me about your task. After a free hour together you’ll get an audit estimate in hours and money.

Most clients come through people I’ve already worked with.