Hi, I'm Martin Korelič.

MobileTransformers is an Android app and framework for on-device fine-tuning and text generation of large language models (LLMs). Built on ONNX Runtime, it supports fully local fine-tuning, federated learning, generation, tool-calling, quantization, and RAG-based generation, with Huggingface transformer models optimized for mobile and edge devices. Deploy your own small or large language models, create on-device AI agents, fine-tune personalized models from private data, and enable retrieval-augmented generation (RAG).

Digital twin platform for real-time environmental monitoring, with a focus on detecting deforestation using satellite data and AI. Integrates segmentation models and leverages Sentinel Hub API for accessing Sentinel-2 imagery. The platform supports easy integration of new models and helps build custom datasets, contributing to research on many remote sensing fields. Made as part of Vseslovenski Arnes HackathON 2025.

On-device semantic location inference system using Wi-Fi SSID sensing and LLMs, enabling privacy-preserving environment context understanding, fully on-device, private and offline. Implemented on Android and paper accepted at ACM EuroSys - EdgeSys 2025 Workshop.

Greenie is a volunteering app for iOS and Android, a centralized platform for all NGOs that create environmental volunteering events. Uses gamification elements, such as scores, leaderboards, streak and progress indicators to activate users to volunteer. Made as part of EU Horizon STEAM 2024 project.

Semantko is a daily Slovenian word puzzle where you uncover the secret word by exploring semantic connections between concepts. Powered by AI and the extensive Slovenian PojemNet network (50,000+ concepts, 500,000+ links), Semantko offers a unique challenge in discovering hidden relationships. Find the links and solve the mystery word.

A machine learning pipeline for detecting Slovenian political bias based on sentiment towards various topics and political parties.

Personal (personality) augmented reality cards with scanning and generating. Made as part of an university course.