Spark sessions 006

The Spark Sessions are a recurring event of the Data Science Initiative at the Faculty of Computer and Information Science. They are designed to facilitate the exchange of ideas and inspiration. The event consists of several snappy (5-7 min) presentations, followed by a casual gathering with refreshments.

When? January 27 2026 at 18:00.

Where? Faculty of Computer and Information Science, Večna pot 113, Ljubljana.

What?

Sunsei: 5-Day Solar Forecasts, Powered by a Novel Regional AI Model - NWPsolarNet
by Marko Rus (Data Scientist, Medius)

NWPsolarNet revolutionizes solar forecasting, moving beyond localized, data-intensive methods. This innovative model, trained on over 300 diverse sites, provides accurate 5-day forecasts for virtually any location, currently focused on Slovenia, without needing historical site-specific data. We’ll detail NWPsolarNet’s technical aspects and introduce Sunsei, a public tool offering powerful forecasting for solar energy optimization.

Dynamic Line Rating in the Electric Power Grid and the Data Science Challenges at Operato
by Frank Amand (CTO, Operato)

What is dynamic line rating and ampacity? Where can data science and ML help us: calculating average/aggregate grid capacity gains, quantifying and reducing uncertainty and operational risk. We’ll also discuss open challenges and opportunities for a Bayesian approach.

Building Secure MCP Infrastructure: From Auto-Generation to Enterprise Hosting
by Luka Gačnik (Software Engineer, Endava)

Model Context Protocol (MCP) is revolutionizing how AI agents interact with tools and data sources. This talk demonstrates MCP Blacksmith, a platform that auto-generates production-ready MCP servers from OpenAPI specifications with built-in authentication and observability. We also preview a hosting platform designed to solve MCP’s security and deployment challenges at scale, enabling data science teams to safely deploy agentic applications with enterprise-grade infrastructure.

Evaluations in the World of Agents
by Jan Hartman (Machine Learning Engineer, Sourcegraph)

As practitioners in data science and machine learning, we’re trained to evaluate every change we make. From offline loss metrics to online A/B tests, we measure everything to ensure data-driven decisions. In the world of AI and agents, however, both input and output spaces are effectively infinite. This talk discusses what this means and presents learnings from hands-on work on agentic systems.

Adapting Glicko-2 for Mario Kart by Blaž Pridgar (Data Scientist, Valira AI)

Our Mario Kart game nights often end in debates about who’s actually the best, so we wanted a fair way to rank players beyond “I won the last race” arguments. Standard rating systems like Elo or Glicko-2 are designed for one-on-one, skill-based games, while Mario Kart is multiplayer, noisy, and partially luck-driven. To finally put the arguments to rest, we built a simple customization of the Glicko-2 system that adapts it to multiplayer races using pairwise comparisons and score-based outcomes.

Want to attend? Attendance is free. Apply here.

Interested in giving a talk at one of the future events? Apply here!