Machine learning engineer working across NLP/LLM systems and computer vision. Experienced with LLM tooling, RAG pipelines, synthetic data generation, vision model training and production Python services, with a practical bias toward building tools that make experimentation and deployment faster.
Brno, Czech Republic
July 2025 - May 2026
Built infrastructure for LLM APIs for fast prompt deployment and development. Worked on tools for prompt iteration, RAG pipelines and backend services in Python. Designed and implemented synthetic data generation pipelines.
Bratislava, Slovakia
July 2022 - June 2025
Implemented back-end for data-related pipelines using microservice architecture in Python, utilizing streaming (Kafka) and OLAP database (ClickHouse). Experimented with Iceberg data format with the goal of improving performance.
Marianka, Slovakia
June 2019 - June 2022
Worked full-stack on an industrial camera tester, including work on a C++ API running on Windows, firmware for embedded MCUs and FPGA for high-speed signal processing, and measurement data processing. Mainly worked on internal tools and partially on new camera products.
MSc, FIT, Faculty of Information Technology
2022 - 2025
Information Technology and Artificial Intelligence (Machine Learning) (EN)
BSc, FEEC, Electronics and Communication Technology
2019 - 2022
Trained and evaluated multiple vision models for ear biometry, such as ViT and ResNet of different sizes, ranging from a few million parameters up to one billion by using LoRA, achieving comparable results with current SOTA. The thesis was presented at Excel@FIT 2025.
Enhancing Entity-Aware Machine Translation with Large Language Models, Retrieval Augmented Generation and Function Calling (ACL Vienna).
A Configurable Toolkit for RAG-enabled LLM Experimentation (COLING Abu Dhabi).
Speaker: Gregor Karetka, Machine Learning Engineer from Slido (part of Cisco). Presented RAGthoven, a toolkit for configurable, reproducible RAG experimentation across execution, evaluation, retrieval strategies, LLM setups and prompts. Recording
Talk about using DuckDB for efficient analytical processing in Python workflows. Covered how columnar execution, local query processing and direct work with common data formats can make big data exploration faster without introducing unnecessary infrastructure. Recording
Achieved 3rd place in the Communication Systems and Network Security category. Honored by sponsor's award of NXP Semiconductors Czech Republic s.r.o.
Contributed to open source software, completing 15 tasks as part of Google Code-in 2016 with main contributions to MovingBlocks/Terasology.
IoT sensor prototype
Nov 2018 - Mar 2019
Developed a prototype IoT sensor for truck detection using a magnetometer processed by STM32. Data was sent back to our server through Slovanet LoRaWAN using an RFM95W LoRa module.
Compotes
Sep 2015 - Aug 2018