Karol Kulasiński

Designed and co-implemented a platform to create, customize, and apply AI agents. The agents disposed of an additional layer of security, thanks to the anonymization service that protects 100% of the data from the client's server. Applied AI agents with the ability to learn on previous tasks and be spawned via Chat, API, or Jira. Deployed the solution to two medium-sized companies.

Led the development process of a scalable exchange platform for personalized ads on Polish TV. Designed logo placement detection AI in live-streamed TV broadcasts. Led the development team of nine, implementing a supply-side platform, end-to-end, from conceptualization and MVP to a scalable production stage. Created an end-to-end pipeline to train the behavioral models based on the data from TV.

Designed and led the implementation of the salad recommendation engine at the production level. Created a BI tool with live-updated sales data and ML forecasting for the CxOs. Built a PoC sales forecasting model based on historical sales and weather forecast data. Improved the legacy ML production models for supply chain forecasting. Managed to improve the processing time by one order of magnitude.

Created a complete training and deploying pipeline for NLP models (BERT) to classify target audience marketing texts. Trained ML models recognizing most similar pictures in terms of content and coloristic that the client provided. Designed and deployed a production-level API for containerized Docker services.

Built an NLP ecosystem for using ChatGPT on the company's private data. Created subMIND, a tool for extracting subconscious information from a large body of text that uses state-of-the-art techniques for entity recognition, graph relations, and visualizations. Built Microsoft Power BI reports for a private Polish university, working directly with the business. Designed an architecture for classifying fake news in social media for the most prominent Polish university, including NLP (BERT) classification, data collection, and overall flow.

Optimized pricing models for online ad auctions using ML tools. Created an entire ML pipeline, including data ingestion, testing, prototyping, error handling, monitoring, and evaluation. Directed the process of product development from the R&D side, including hypothesis testing and handling client feedback

Created Tableau reports identifying fraudulent behavior of employees. Built a fully automated quality assurance system for data ingestion. Designed a Twitter fake news detector front end for data visualization.

Carried out state-of-the-art research using molecular dynamics and Monte Carlo simulations on nanoscopic materials. Published three technical papers in a highly respected scientific journal. Created, simulated, and interpreted numerical simulations with over 10^7 degrees of freedom.

Carried out numerical simulations that resulted in models further used by other team members. Published nine technical papers in top-ranked journals as the first author. Contributed to the physical chemistry field by explaining the water adsorption-related phenomena in cellulose.

Created a numerical model of the secondary loop of the BWR nuclear reactor under the direction of Professor J. Ragusa. Applied the Monte Carlo method for sensitivity analysis of numerical coefficients in different equation functions of the state. Expanded the lab's Python library for carrying out finite element method simulations.