- Signal processing.- Multi-objective optimisation (ETA, fuel consumption).- Ship routing.- Vessel fuel optimisation by speed-fuel curve fitting.

• Developed and implemented grinding and recovery quality predictive models to meet user (metallurgists) requirements, reducing controlled parameters by 20% and state parameters by 70%. Improved model reliability and accuracy, reducing MAE by 2x. • Led a team of 4, and initiated the Data Strategy.Result: 3% increase in copper plant productivity.

3-months heritage sites project in university, customer — bank• EDA with visualisation on folium• Made special metric for availability of certain categories of commercial establishments, so based on it Got simple static recommendations for heritage sites commercial using• Used OpenStreetMap

• Continued using weather data for analysis and feature engineering• Created anomaly detection algorithm for vessels based on geohash — it was so good that we decided to start new commercial product based on it• Another participating in decarbonisation hackathon, without a place at this time, but I got experience with reinforcement learning

Joined DS team and brought a lot of creativity and new:• Started using folium for visualisation• Found bottleneck in data preparation• Created new reliability model for post processingAlso had general experience with time series data prediction:• Used Keras for model fitting (LSTM)• Created new metric for crossing paths• Added new features which improved model by 10-50%, depends on special sub area.More work with weather data analysis:• Was responsible for this type of data in DS team• Recode weather loader from Java to Python• Found critical problem in one type of data, so another department didn’t download crushed TBs of data• Explored and successfully used tides calculator

I worked on Weather project with Senior Meteorologist.• Used open NOAA data• Tried classic algorithm based on locator and masks (Objective fronts, Hewson, 1998)• Improved masking method by adding new parameters and layers• Added post processing (noise removing): clustering (DBSCAN) + shapely• Developed plugin for QGIS 3, where Meteorologist can add data for exact date and then he will get atmospheric fronts (Polylines geometry object) for whole world (except the land).• Error of calculator is about 100-120km

Scientific research project for heat accumulators (HA) optimising • Combined math models of HAs, optimised parameters (used Nelder-Mid).• Created GUI for engineers on Python: run calculations and optimisation.