Freelance Data Scientist
2018 - PRESENTFreelance Data Scientist- Gathered and presented data from coffee shop registers and the derived customer behavior patterns to enable the marketing team of the beverage producer to make better decisions on how, when, and where to invest the marketing budget.
- Developed a suite of spend classification models using R language (data.table, ggplot2, xgboost packages), NLP techniques and XGBoost classifier, used AWS Lambda and AWS API Gateway for production deployment.
- Designed an expert system to enable the client to deliver expert procurement knowledge on creating procurement strategies for his customers.
- Wrote extensive documentation of the expert system solution to serve as a basis for patent application.
- Developed a reporting database based on PostgreSQL, using Power BI as frontend. Implemented a data pipeline using R language (tidyverse, jsonlite, httr packages) to integrate with clients Square and Brushfire accounts using Square and Brushfire APIs. The PowerBI dashboards covered business sales, inventory and labor business areas.
- Authored a technical whitepaper on an edge-based machine learning solution for a client.
- Delivered a "Data visualization 101" workshop on multiple IT conferences and meetups. The workshop focused on basic data visualization principles - from how human visual cognition works, to basic data visualization forms and most frequent mistakes. There was also an emphasis on creating effective dashboards.
Technologies: Amazon API Gateway, AWS Lambda, Apache Spark, Hadoop, RStudio Shiny, DataTables, sparklyr, Purrr, Tibble, Readr, Ggplot2, Dplyr, Tidyverse, RData Scientist
2017 - 2018Hrvatske telekomunikacije inc., Zagreb, Croatia – part of Deutsche Telekom- Served as a member of an international analytics team of Deutsche Telekom, working remotely from Croatia, with the team manager in Germany. I've used Oracle SQL on the Oracle 12c data warehouse as a data source.
- Fixed lines churn prediction model enabled early detection of customers with potential to terminate the service, enabling preventive retention actions. I've used Oracle SQL on the Oracle 12c data warehouse as a data source and SPSS Modeler for modeling and deployment to production.
- Improved households detection significantly increased the potential base of customer households, necessary for offering the companies' flagship product. I've used Oracle SQL on the Oracle 12c data warehouse and Hive SQL on a Cloudera big data platform as a data source, H2O for modeling and R (data.table, H2O, cronR, ggplot2 packages) for additional data preparation, deployment to production and monitoring.
- Developed propensity models for key products significantly increased the conversion rate. I've used Oracle SQL on the Oracle 12c data warehouse and Hive SQL on a Cloudera big data platform as a data source, H2O for modeling and R (data.table, H2O, cronR, ggplot2 packages) for additional data preparation, deployment to production and monitoring.
Technologies: SPSS Modeler, SQL, Apache Hive, Big Data, Cloudera, Oracle SQL, Machine Learning, DataTables, Cron, H20, Ggplot2, RData Scientist
2008 - 2017Vipnet LLC, Zagreb, Croatia – part of América Móvil- Built a recommender engine generating individualized product suggestions for each business customer, by combining internal and third-party data on business customers. I've used Oracle SQL on an Oracle 12c data warehouse as a data source.
- Estimated the potential for fixed network expansion with pinpoint accuracy on individual address level for the entire territory of Croatia by combining public and internal company data. It enabled optimal allocation of investment in the fixed network – to areas with the most commercial potential, and lowest construction costs. I've used Oracle SQL on an Oracle 12c data warehouse as a data source.
- Trained a model estimating the likelihood a customer owns a competitor subscription by combining market research data with internal data. It provided a potential base for cross-sell/up-sell activities. I've used Oracle SQL on an Oracle 12c data warehouse as a data source, SAS for data preparation and deployment to production, SAS Enterprise Miner for modeling.
- Analyzed customer recharge behavior by creating a recharge based segmentation. The segmentation enabled introduction of new voucher denominations more suited to customer needs. I've used Oracle SQL on an Oracle 12c data warehouse as a data source, SAS for data preparation and deployment to production, SAS Enterprise Miner for modeling.
- Developed a model predicting which customers are most likely to buy data options. It enabled optimal customer targeting when offering data options. I've used Oracle SQL on an Oracle 12c data warehouse as a data source, SAS for data preparation and deployment to production, SAS Enterprise Miner for modeling.
- Analyzed the purchase behavior of small businesses by applying market basket analysis to purchase transaction data. It provided new insights usable by sales. I've used Oracle SQL on an Oracle 11g data warehouse as a data source, SAS for data preparation and deployment to production, SAS Enterprise Miner for modeling.
- Created models predicting churn for the small business segment. They enabled early detection of customers with potential to terminate the service, enabling preventive retention actions. I've used Oracle SQL on an Oracle 12c data warehouse as a data source, SAS for data preparation and deployment to production, SAS Enterprise Miner for modeling.
- Collaborated with the data warehouse team to redesign the data science data mart. We were engaged in the definition of data sources, data transformations, and database table formats. Following the implementation, we did intensive data quality testing. The resulting data mart was much more suited to our needs and had traceable data sources, which helped to quickly resolve data quality issues. I've used Oracle SQL on an Oracle 11g data warehouse as a data source.
- During the DWH redesign project, I recognized the business need for a unique customer data set. I've compiled a detailed specification containing complex rules on data processing and data quality improvements. In the process, I profiled two relevant source systems which contained customer data. The resulting unique customer data set is used for company-wide reporting, CRM campaigning and has enabled a tenure-based customer loyalty program. I've used Oracle SQL on an Oracle 11g data warehouse staging area as a data source.
- Implemented an e-bill affinity prediction model, which predicted which residential customers are most likely to switch to e-bills. It enabled the billing department to speed up the adoption of e-bills. I've used Oracle SQL on an Oracle 11g data warehouse as a data source, SAS for data preparation and deployment to production, SAS Enterprise Miner for modeling.
Technologies: SPSS Modeler, SAS Enterprise Miner, SAS, Oracle SQLBusiness Intelligence Developer
2007 - 2008SoftPro Tetral LLC, Zagreb, Croatia- Contributed to development work on CubePlayer application, an OLAP client for Analysis Services 2000/2005 using VB.NET 2.0, MDX and ComponentOne for .NET 2.0.
- Introduced ClickOnce deployment, Subversion source control and Trac issue tracker into the CubePlayer development project.
Technologies: Trac, Subversion (SVN), .NET, ComponentOne, Visual Basic .NET (VB.NET), ADOMD.NET, MDXTeam Lead
2007 - 2007Ekobit LLC, Zagreb, Croatia- Developed Taxman, a tax return application targeted to the German consumer market and developed for a German client company Lexware GmbH. I've used C# 2.0, NET Framework 2.0, SQL Server 2000, MS Access 2000 and C++/MFC.
- Lead a team working remotely on full stack development of Taxman.
Technologies: Microsoft Foundation Class Library (MFC), C++, Microsoft Access, SQL Server 2000, .NET, C#Software engineer
2004 - 2007Ekobit LLC, Zagreb, Croatia- Developed MAWIS, an ERP system used in the waste disposal industry developed for a German client, MOBA AG. Work involved maintenance and implementation of new functionality. I've used C++/MFC and SQL Server 2000.
- Built MAWIS-online, a lightweight web-frontend for the MAWIS ERP system using C# 2.0, .NET Framework 2.0, SQL Server 2000.
- Created MAWIS.NET, a framework for import/export of data to/from MAWIS ERP system using C# 2.0, .NET Framework 2.0 and SQL Server 2000.
- Worked remotely on all above mentioned software development projects.
Technologies: C++, Microsoft Foundation Class Library (MFC), SQL Server 2000, .NET, C#Software engineer
2002 - 2004Okit LLC, Zagreb, Croatia- Developed ZAD3-online, a web application used for registration and tracking of failures in the low-voltage power grid developed for a Croatian power utility company using C# 1.0, ASP.NET 1.1 and Oracle 9i.
- Built ZAD3, a Windows application used for registration and tracking of failures in the low-voltage power grid developed for Croatian power utility company, using C++/MFC and MS Access 2000.
- Programmed ZAD1, a Windows application used for registration and tracking of failures in the high-voltage and medium-voltage power grids developed for Croatian power company using C++/MFC, MS Access 2000 and Oracle 8i.
Technologies: Microsoft Access, Microsoft Foundation Class Library (MFC), C++, Oracle9i, ASP.NET, C#