Nazar Barabash

The fintech company from the Bay area built its own computation engine to serve performance management reports. However, in time, the amount of data has grown significantly, and the user wanted to be able to drill down to the lowest levels of the reports. The custom execution engine was not able to handle such a request.
I built a new Spark-based engine by moving data from Neo4j and RDBMS to parquet files. The new engine was able to process more data and provide more granular reports.
Switching to the new execution engine allowed my client to keep ongoing contracts, onboard new clients, and later sell the business with a good position on the market.

The biggest Canadian retailer decided to extract new insights and business value on top of siloed data.
I led a team to design and implement a comprehensive DSL-based ingestion framework. The framework provided out of the box features like extensibility and code reusability, unified logging, integration testing (in a small Docker-based cluster), and single command deployment.
Using the implemented framework, our team set up about 1,000 ingestion pipelines, data deduplication, and aggregation jobs in a few months. We provided a solid data set to the data science team.

I was engaged on behalf of Canadian retailer to design and develop a proof-of-concept search engine solution for support of data discovery activities over sub-set of the client’s existing enterprise data warehouse.
By itself, data was collected in Data Lake and the analytics built on top of it did not provide many insights to business in time because of difficulties in accessing data. Proper reports and research requests take weeks.
With the help of a UI engineer, implemented a single page application where businesses can easily query for high-level data with human-readable requests like "top sales in 2020 by province." Direct match search also proposed relevant results for similar dimensionality.
The business was able to query data on demand without requests to BA teams.