Projects

An overview of selected academic and personal projects, with demos, papers, videos, and public project pages collected in one place.

Academic Projects

Illustration for the LAD project

LAD: LoRA-Adapted Diffusion

Academic

LAD is a diffusion style LLM. The project includes code, interactive demos, the EMNLP paper and an explanation video.

Illustration for the SHAP Sum project

SHAP Sum

Academic

An interactive Streamlit project around an abstract presented at MEMTAB, focused on traceability of LLM summarization based on SHAP.

Illustration for the Inspect-SR project

Auto-INSPECT-SR

Academic

Auto-INSPECT-SR is a demo for automated trustworthiness screening of randomized controlled trials, based on the Inspect-SR tool described in the paper by Wilkinson et al.

Screenshot for the EHR-to-struct project

EHR-to-struct

Academic

A demo showing how LLMs can extract structured information from simulated Dutch EHRs, preserve traceability through extracted spans, and map the output into the OMOP common data model.

Personal Projects

Illustration for the TiniTopo project

TiniTopo

Personal

TiniTopo is a high-resolution terrain viewer that blends open-source national topography datasets with AI-enhanced worldwide elevation coverage into a unified global map. It supports detailed terrain exploration, OpenStreetMap building overlays, and model export for 3D printing.

Illustration for the Rise or Ruin project

Rise or Ruin

Personal

A turn-based strategy board game for 2 to 6 players, playable against a local AI opponent, directly in the browser. It combines empire building, resource management, and unit battles. This was created as an experiment to test the performance of different LLMs in never-before-seen games. Uniquely, it allows LLMs to play eachother without any user interference.

Illustration for the Homie project

Homie

Personal

Homie is an always-on Raspberry Pi voice assistant for home use. It listens for a wake word, records commands, transcribes speech, generates responses with local or cloud models, and speaks the answer back through connected audio devices.

More