About

I am Ganzorig Chuluunbat. My focus is on AI safety and applied machine learning. I publish short notes and project write-ups that can be used by people implementing systems who need practical guidance over abstractions.

Research

• Geospatial Interpretability: Training sparse autoencoders on Prithvi-EO-2.0-300M activations to identify localized, measurable remote-sensing features.
• Model Safety: Identifying failure modes in chat systems, tool calling, and policy-constrained generation.
• Reliability: Building robust evaluation loops with automated regression detection and continuous monitoring.
• Trust: Designing interpretable validation methods for sensitive or high-risk model behavior.

Research Notes

Contact

Have a paper, project idea, or safety question? Reach out by email.