Data reliability you can see
Criticality tiers, freshness and uptime targets, and a test that catches seconds-vs-minutes unit bugs.
dbt project on GitHub →I’m Artur Kochetkov. I build Snowflake and dbt platforms, reliability systems and LLM tooling, including a program that cut six figures a month from a marketplace’s warehouse bill.
The stack I work in every day
A query that costs too much. A table that’s late again. A feature that needs tracking before it ships. I make each of those visible, owned and fixed, so the team can get back to building.
Price every query by data scanned and credits, then fix the expensive ones first.
Freshness and uptime targets by tier, so on-call knows what’s late before analysts do.
From data model to App Store, including the backend, the site and the launch.
Every legacy query got a price tag: data scanned, runtime and Snowflake credits. An LLM grounded in table metadata and that cost profile proposed a rewrite, and each rewrite was proven to return the same results before anyone reviewed it.
4,000+ SQL files across marketing and analytics pipelines. Rule one: the numbers must not change.
File names and per-query figures are illustrative. Program totals below are real.
Execution time, data scanned and credits from query history, so every file has a cost before anyone touches it.
An LLM, given the query, its table metadata and its cost profile, proposes SQL that keeps the original intent.
Before and after results are compared, savings simulated, and each file gets a diff, an estimate and a priority.
Criticality tiers, freshness and uptime targets, and a test that catches seconds-vs-minutes unit bugs.
dbt project on GitHub →An AI-assisted workflow that turns a requirement into events, a field schema and a KPI check before engineering builds.
Ask me how it works →At AliExpress my team raised new-buyer LTV from social by 38%. This open project shows the method on GA4-style data.
social-commerce-ltv on GitHub →First, last, linear, time decay, position based and a Markov chain that learns from journeys that didn’t convert.
See the models →Find the behaviour that predicts value, then test a nudge toward it and read it out with variance reduction.
Read the experiment →Led a team of five moving Braze and lifecycle feeds off legacy SQL jobs onto tested, incremental dbt models.
See my experience →I treat warehouse spend like latency: measured per query and per team, with a budget.
Contracts, generic and singular tests, and parity checks so incremental output equals a rebuild.
Every published table has a tier, an owner and a freshness target anyone can check.
AI tooling grounded in metadata and hard constraints, with a human reviewing the diff.
Hiring, roadmaps, code review and standards, as a Director, a Head and a Lead.
Registration and hire count the same in product and finance, because the events are configured once.
Catalog metadata on every published table: dataset, owner and grain, without asking.
Social, lifecycle, attribution and live shopping: I’ve owned that data from first touch to purchase.
I ship my own products, so I know what “done” costs on the other side of the dashboard.
Pillwave reminds people to take their medication. You snap the pill in your hand to check in, and family members get a quiet “✅ took the pills.” I designed it, built it and run its launch.
The whole stack, solo. Expo and React Native on the phone, SQLite offline-first storage, Supabase sync with per-row access rules, and a static site on Cloudflare.
Privacy is the product. Check-in photos never leave the phone. Without a linked family member, nothing does.
An AI marketing team. 23 Claude agents draft posts, check them against medical-claim rules and queue them for my approval.
I work inside your warehouse, orchestrator and BI tool. No rip-and-replace, no new platform to buy.
Daily drivers
Also at home in
Certified in Data Engineering, Big Data and ML on Google Cloud, and in ETL pipelines with Airflow and Kafka.

Artur Kochetkov
Houston, TX · Remote, US
B.S. Business Administration, CSU Channel Islands
I’ve led teams of 7 and 10 as a Director and a Head, and I still write the SQL. I care about the parts that make data trustworthy: tests, SLAs, clear ownership and a warehouse bill nobody is afraid of.
LLM SQL optimization across 4,000+ queries. Led marketing data engineering with 5 engineers, moving lifecycle feeds onto dbt, and built the platform reliability view with SLA tiers.
Led a team of 7. Rebuilt the BI platform for full-funnel marketing visibility and built predictive models on marketing spend.
Led a team of 10 and a real-time performance portal. +38% new-buyer LTV from social channels.
Forecasting, OLAP reporting and Tableau. Opened the company’s first profitable Moscow store in 19 years.
For hiring managers and recruiters.
For teams with a data problem to scope.
Consulting rates depend on scope and are shared after the intro call.
Cost visibility works best when it is proactive, not a monthly surprise.Lesson from the LLM SQL optimization program
Go deeper
Yes. I’m based in Houston, TX and work remotely across US time zones. I’m happy to be on site in Houston, and I travel for planning weeks.
No. I’m authorized to work in the United States and don’t need sponsorship.
I take a small number of scoped projects that don’t conflict with my employer. If I join your team full-time, you come first.
Mostly Snowflake and dbt, orchestrated with Airflow, reported in Looker or Tableau. I’ve also worked with BigQuery and the GA4 export, and my public projects run on DuckDB.
Yes. social-commerce-ltv and marketing-data-reliability are public, with tests and CI. Employer work is shown as anonymized case studies.
By judging social channels on what buyers were worth months later, not on the first order, then segmenting early behaviour and shaping campaigns toward repeat purchase. The social-commerce-ltv project shows the same method on synthetic data.
Daily, with guardrails: grounded in real metadata, held to hard constraints like “results must match”, and reviewed by a person before anything ships.
Pillwave is in pre-launch for iPhone with a public waitlist at pillwave.app. I built the app, the backend, the site and its marketing system myself.
Usually within a day. The quickest route is to book a call; you’ll see my open times in your own time zone.
A role to fill, or a warehouse bill nobody can explain. Either way, 30 minutes is enough to know if I can help.
Free intro call · Central Time · Usually replies within a day