2026-02-10

Why ML Models Should Fail Gracefully

Most evaluation frameworks optimize for peak accuracy. But what happens when inputs fall outside the training distribution? I've been thinking about failure-first metrics and what they reveal about model robustness.

2026-01-25

Liquid Neural Networks: First Impressions

After months of working with LNNs for ECG classification at IIT Hyderabad, here are my observations on their interpretability advantages and the trade-offs compared to traditional architectures.

2026-01-08

Building for 100 Languages: Lessons from CVPR

Our paper on evaluating LLMs across 100 culturally diverse languages was accepted at CVPR 2025. Here's what we learned about the blind spots in multilingual AI evaluation.