Vasilii Molchanov

Analytics / Data Engineer

Four years in data, the last three on incremental ELT on MPP and lakehouse platforms (Iceberg, CDC replicas): data quality checks, step-by-step reconciliations and alerts that stop a load when a check fails. Domain so far: operational, credit and liquidity risk and provisions in banking.

End to end

Operational risk in banking, from source to dashboard, shown on synthetic data

Spring 2026

Operational Risk Overview

Monthly risk report for the board: step-by-step SQL models, automated mapping of business processes to P&L lines, a ClickHouse mart and custom Superset charts.

SQLGreenplumClickHouseApache SupersetECharts

Summer 2026

Operational Risk Internal

Analyst workbench on the incident database: data moved from the Iceberg lakehouse into ClickHouse and reconciled with the board report month by month, behind 14-month trends, variance heatmaps and drill-down.

SQLTrinoIcebergClickHouseApache SupersetECharts

Research

Master's project at HSE University

Autumn 2025

Brand Mentions in Russian Lyrics

Corpus of 2,967 chart songs from 1991 to 2025: LLMs extract brand mentions from lyrics, and a dashboard tracks them by year, genre, category and origin.

PythonpandasLLM APIsFlaskChart.js

Personal

Summer 2025

Diet Matrix

High-protein meal planner generated by Python from CSV data: swappable dishes, ingredient-level macros, a meal tracker and a priced shopping list.

PythonJavaScriptHTMLWeb scraping

Spring 2025

Workout Calendar

Training log modelled in dbt on DuckDB: CSV sources, staging, intermediate and mart layers, with the 42-day scoring ported from Python to SQL window functions. 45 tests run on every build, one of them checking each session's score against the Python original.

SQLdbtDuckDBPython

Pipelines

Banking, 2023 to 2026. The code stays with the employer, so it is described, not shown

  • Ingestion

    Daily incremental load into the loss event database

    Picks up new and still-open records from ledger and CRM replicas, matches losses to customer compensations and writes events, losses and recoveries. All data quality checks run as one gate before the write: any failure stops the refresh and posts an alert.

  • Reconciliation

    Completeness check for the credit fraud pipeline

    Rebuilt the full history from the upstream risk mart and traced every contract through the pipeline as a funnel: about 95,000 contracts in, each filter step counted and explained, 2,200 loaded. The gaps it exposed became a backfill of missed losses.

  • Schema change

    Pipelines moved to new sources with history intact

    Upstream retired the anti-fraud source tables and split the ledger into a six-month hot table and a frozen archive. The affected pipelines were moved to the new sources, old and new outputs compared, and the records missed since 2023 backfilled.

  • Impact analysis

    Reference data migration

    A shared dictionary gained duplicate codes that multiplied rows on every join. Dependencies were traced through the team's scheduled code, and the blocked loaders were switched to a deduplicated copy with a row-count alert against the source.

  • Deduplication

    Adjustments attached to the incidents that caused them

    Customer adjustments caused by IT incidents were booked as generic adjustments, apart from the incidents. A cross-process key built from the ledger posting now attaches them to their incident; reloading history since 2024 added about 1,300 incident events and cancelled about 3,000 conflicting ones.

  • Automation

    Regulatory assessment from spreadsheet to monthly pipeline

    The bank's economic position score under the central bank's methodology was calculated by hand. A monthly pipeline now computes the ratios from regulatory reporting marts, was checked against past submissions, and emails a draft for review.

Toolbox

Storage
MPP warehouse (Greenplum), lakehouse on Apache Iceberg (Trino), ClickHouse, PostgreSQL
Pipelines
SQL, Python, incremental ELT from CDC replicas, dimensional models (star schema) layered from staging to marts, backfills, DAG orchestration; dbt, Git and GitHub Actions in personal projects
Quality
Data quality gates, reconciliations, alerting, data catalogue, code review
BI development
Apache Superset, Tableau, Power BI, ECharts, Jinja, Handlebars, JavaScript
AI tools
Claude Code and Antigravity in everyday development, LLM APIs for text extraction

Background

2023 to now
Analytics Engineer, T-Bank (Tinkoff)
2023
HR data analyst intern, Ozon Tech
2022 to 2023
Research data analyst, HSE Institute of Demography
2023 to 2026
Master's in Digital Humanities, HSE University: text analysis and NER
2019 to 2023
Bachelor's in Sociology, HSE University: statistics and data analysis