Maaz Karim

I'm Maaz Karim

- Building video AI systems.

I learn, tinker, and ship better models.

Most of my work lives in the gap between model experiments and usable software training and tuning models, shaping data pipelines, evaluating tradeoffs, and getting systems into a state that is reliable enough to run repeatedly.

Recently

  • Building production ML systems for long-form video understanding.
  • Working on automated chapterization for live news streams.
  • Exploring VLMs, long-video reasoning, and multimodal evaluation.
  • Working toward stronger research projects in video understanding, embodied AI, and agentic systems.