projects

Undertakings, past and present, mostly related to research and open-source software.

EU-Agent-Bench

Expert-curated, verifiable benchmark for LLMs in an EU context

Policy optimization over SAE features

Using RL to steer sparse autoencoder features as an alternative to RLHF

'Aha moments' during reinforcement learning

Investigating phase transitions during Group Relative Policy Optimization

AI Agent for the State Statistical Office

Agentic system for policymakers and journalists to extract insights from national statistics

Macedonian multimodal AI infrastructure

Building open-source LLM infrastructure for the Macedonian language

Synglot

A Python library for translating and generating large datasets

Compositional reasoning

Benchmarking LLM generalization with tunable difficulty axes

Applied vibes research

Exploring cultural insights through embedding geometry

Mitochondria under distress

Investigating mitochondrial dysfunction using solid-state NMR

ESA ATHENA mission

Investigating calibration anomalies for the X-ray space telescope

service

Reviewing:

AI safety:

  • overseeing the AI Safety Initiative Groningen Research Hub, setting up research priorities, conceiving topics, and securing partnerships.

Macedonia:

  • Mentored 7 students in ML engineering and product design
  • Advising Mostovi on matters of digitalization and strategy.
  • overseeing finance at the UWC Macedonia, performing financial needs assessment; allocated several million $ worth of scholarships over 4 years