Research Interests

Developing evaluation frameworks that measure AI creativity beyond simple novelty. Current work focuses on LLM-as-a-Judge methods, testing the limits of open-source judge models, and building reward models that capture the gap between what's creative and what's actually enjoyable to read.

Exploring how multiple LLM agents can collaborate on creative tasks. Interested in emergent behaviors when agents with different specializations (critic, writer, editor) negotiate on output quality — and whether collective judgment outperforms single-model evaluation.

Extending the AI creativity evaluation framework from text to images. The challenge: visual creativity has different dimensions than textual creativity — spatial composition, style coherence, and emotional impact need different measurement approaches.

Working toward building a native LLM for the Khmer language. Khmer is severely under-represented in existing training corpora, and its unique script and morphology pose specific tokenization and embedding challenges that don't transfer well from high-resource language approaches.