Zivan Karaman

Working for clients in Europe and North America, I help them transform complex data into valuable knowledge to support innovative projects.

I provide research, large data set analysis, and software design and development services.

Some recent and/or ongoing projects:

- Modification and optimization of a large automated geospatial data processing pipeline to replace obsolete R packages (rgdal, rgeos) with recent libraries (sf, terra).

- Analysis of large datasets from simulations for the development of national electricity transmission networks and visualization of results on an interactive map.

- Development of client software for geospatial and remote sensing APIs.

- Mixed model analysis of large-scale experiments on biotechnological traits.

I led a team of scientists and computer scientists working on the design and deployment of innovative digital tools to help farmers get the most out of their fields and crops.

Geospatial data was at the heart of the system: soil and topographic maps, satellite/aerial imagery, yield maps, etc. New models and algorithms were developed by combining crop growth modeling and machine learning approaches, adapted to geospatial data and using geostatistical tools.

A dedicated Web platform and associated mobile application, available in more than a dozen languages, were developed. The backend was hosted in the cloud, dockerized, based on the PostgreSQL / PostGIS / Geoserver open source stack, using Shiny/Leaflet dashboards for reporting, and interacting directly with internally developed R packages implementing agronomic models. The system was automated, robust, industrial-grade, providing 24/7 service. This enabled the large-scale deployment of innovative digital tools, serving thousands of users in many countries.

The focus was on association studies and genomic selection, due to the availability of affordable, high-throughput DNA analysis techniques. The main tasks were the preprocessing of phenotypic and genotypic data, the construction and validation of models, and the prediction of individual performance. Bayesian methods and machine learning methods, such as ridge regression, lasso, and random forests were used, as well as computer simulations of breeding programs. The tools were mainly implemented through public or internally developed R libraries, and were made available to users via a dedicated web portal. Particular attention was paid to quality control to ensure the reliability and robustness of the models produced. The strong automation of the process was achieved and the system enabled the widespread application of molecular tools in breeding programs and the introduction of new variety creation schemes.