Systems currently running in production. I don't name the companies — I describe how the systems work.
01
Product feed robot
Rebuilds a product feed for ad platforms every day and tracks catalogue changes.
- Parses the site catalogue, builds a YML feed and publishes it to object storage under a permanent URL.
- Diffs against the previous day: new and missing items, price changes — the history is kept in the database.
- Runs on a cloud timer; a run report is delivered to a messenger.
- A separate module generates smart banner imagery and tracks the actual generation cost.
Source on GitHub →PythonYandex Cloud FunctionsPostgreSQLS3
02
End-to-end analytics
Links ad spend to leads and deals from the CRM to measure cost per lead and cost per sale.
- Daily ETL jobs pull campaign statistics and CRM deals into a shared database.
- A schema per client, plus a reconciliation job that backfills missing data.
- Lead qualification rules and campaign mapping live in configuration, not in code.
- Plan-versus-actual and channel reports are computed on top of the database.
PythonPostgreSQLETLYandex Direct APIBitrix24 API
03
Agent system for marketing work
A set of agents and skills that carry repetitive advertising work end to end.
- Skills describe a full task pipeline: keyword research, cleanup, clustering, campaign assembly.
- External data sources are connected through MCP.
- Ad copy and image generation with limit and cost checks.
- Shared safety rules: secrets outside the code, explicit confirmation before touching live ad accounts.
LLMMCPPythonOpenRouter