Viktoriia – SQL, Airflow, CI/CD

The data foundation of a B2B marketing platform for 300+ e-commerce clients. It covers client billing data, data ingestion from client systems, data transformation models, and the migration of transactional data processing from Google Cloud (GCP) to AWS.

• Owned the billing data platform end-to-end, reconciling Snowplow, Klaviyo, and email service provider (ESP) event streams into one source of truth. This eliminated recurring billing discrepancies across 300+ clients;
• Re-platformed a 200-client Airbyte deployment from a single EC2 server to a multi-tenant Kubernetes cluster on EKS (Terraform, Karpenter, KEDA). This eliminated ~96% idle capacity and raised the ceiling from 200 to 1,000 concurrent clients;
• Chose autoscaling on Kubernetes (Karpenter adds or removes servers, KEDA adds or removes workers based on queue size) over one always-on server, so the company pays only for capacity actually in use;
• Re-architected 12 dbt model families from full rebuilds to incremental merges, which process only new or changed data. Pipeline runtime dropped by 40% with row-level parity, meaning the results were identical;
• Designed and built the AWS landing zone (VPCs, EKS, RDS, Prometheus/Grafana) for moving transactional ETL from BigQuery to S3 Iceberg + Glue. It was delivered as reusable Terraform modules, with every change tested on staging before production;
• Set CI/CD and infrastructure standards for the data team, using GitHub Actions pipelines for Flink jobs, Airflow DAGs, dbt models, and Docker image builds;
• Used Claude Code to speed up delivery;
• Worked directly with account managers and finance to define billing rules and onboarding requirements, turning unclear requests into production pipelines.

A real-time marketing data platform for e-commerce brands. It recognizes website visitors, links their activity to a single shopper profile, enriches profiles with additional attributes, and predicts which shoppers are worth contacting. It serves 300+ online stores, 500M+ shopper profiles, and billions of behavioral events per day.

• Chose a streaming architecture using Flink and Kafka over the existing batch process, enabling shopper events to be matched in real time rather than hours later and supporting instant personalization and outreach;
• Built a real-time identity-resolution pipeline using Flink, Kafka, Redis, and Neo4j to link visitor events into unified shopper profiles. Replaced the existing Google Tag Manager-based batch process and reduced event resolution time to under one second;
• Selected Neo4j as the graph database for identity resolution, enabling efficient traversal of connections between shoppers’ emails, devices, and cookies;
• Developed core identity-matching components, including identity waterfall prioritization to select the most reliable identifier when multiple identifiers conflicted. (IP-based matching for visitors without an identifier; Form-submission event capture; Cart-token enrichment)
• Built a real-time predictive scoring engine processing behavioral signals across 500M+ shopper profiles and billions of daily events. Reduced outreach volume by 25% while maintaining the same revenue;
• Designed and delivered a self-service client onboarding flow using Cloud Run, Airflow, Slack, and a web UI, reducing onboarding time from 15 days to 4–8 hours;
• Developed an attribute-enrichment platform for 300+ client accounts using Pub/Sub, Cloud Functions, MongoDB, and Retool, with an interactive preview that returned results in under 30 seconds;
• Used Claude Code to accelerate development and delivery.

Internal analytics tools for a performance marketing team that runs paid advertising for 100+ e-commerce brands. They include an AI assistant that analyzes daily campaign performance, a set of optimization models, and an automated system that catches mistakes in ad campaign settings.

• Built the data backend for an AI analytics agent, enabling LLM-driven daily performance analysis for 100+ clients;
• Developed a FastAPI service exposing marketing optimization models for budget reallocation, spend pacing, creative fatigue, seasonality, and geographic saturation;
• Designed an automated internal-controls engine for media buying using declarative SQL rules to detect budget, bidding, and targeting violations. Ensured rules ran only when source data was fresh and integrated Slack alerts with a Vue 3 triage dashboard;
• Used Claude Code to accelerate development and delivery;
• Worked directly with non-technical teams to understand their needs and deliver solutions.

A data platform that collects and loads daily data from 100+ e-commerce clients into one database for marketing analytics. Each client has its own data format, so the platform maps, validates, and monitors every client’s data separately.

• Built and owned a multi-tenant ETL pipeline from GCS to AlloyDB serving 100+ clients. It has client-specific mappings, safe re-runs without duplicate data, and file-level tracking of what was loaded and when;
• Reduced the largest client’s daily load runtime by 35% by rewriting the merge logic and tuning batch sizes;
• Built a data quality and alerting system with row-level validation and detection of unexpected changes in source data structure. It automatically creates Slack and ClickUp tickets, catching 95% of issues before production and cutting average fix time from 4 hours to 20 minutes;
• Migrated legacy scripts from PythonAnywhere to GCP as versioned services deployed through CI/CD.

An internal data and reporting platform for an IT services company. It combines recruiting and HR data from internal and external sources into performance dashboards for management.

• Designed and maintained Python and SQL Server ETL pipelines that combine recruiting and HR data from internal and external sources;
• Optimized Microsoft SQL Server queries and added database monitoring, improving report load time by 25%;
• Refactored the legacy codebase to improve maintainability and reduce code duplication;
• Built Power BI dashboards for performance metrics, translating stakeholder requirements into technical solutions;
• Worked directly with non-technical teams.