Reliability Before Leaderboard
In deployment-facing perception tasks, predictable behavior under failure can matter more than benchmark margin gains.
Read NoteResearch Notes
Short reflections on robust AI, temporal reasoning, and student-to-researcher growth.
Why Temporal Context Matters
Weak-target perception improves when sequence-level evidence is modeled explicitly rather than frame-by-frame only.
Read NoteFrom Coursework to Research Confidence
Reproducible experiment design and clear uncertainty reporting are key to transitioning into research-quality work.
Read Note