Sneha Hassan

I'm a software and AI/ML engineer. I build systems that learn: most recently benchmarking large language models on biological reasoning at Carnegie Mellon's xulab, and before that shipping production ML on electric scooters at Ather Energy, where a pothole-detection model I built reached 92% accuracy on real roads.

I like problems with weird shapes: an AI tutor that refuses to give answers, a VR rehab platform, the world's smallest bipedal robot. Equal parts tech nerd and people person; I turn chaotic ideas into shipped products, preferably before lunch.

Work

Selected projects, from LLM research to production ML. See all projects or my experience.

BioSemantic Bridge: benchmarking bio-domain LLMs with a multi-agent ensemble decoderSocratic LLM Tutor: pedagogical AI that guides with tiered hintsPothole Detection: accelerometer ML deployed across Ather's rider baseRecoVR: VR assistive therapy, incubated at CMU

Writing

More on the blog, older essays on the writing page, some loose scribbles, and the full archive on Medium.

Contact

The fastest way to reach me is snehahassan7920@gmail.com. Here's my resume (other versions); I'm also on GitHub, LinkedIn, and Medium. More about the human behind the resume.

Currently

  • My tech-fluencer era
  • Teaching three models to vote on bio-imaging questions at CMU's xulab.
  • Validated Agent Memory is a research prototype that tests whether AI agents improve more reliably when they only remember lessons that have been checked for correctness.
  • RecoVRBUILDING
    VR physiotherapy, incubated at Carnegie Mellon.

P.S. Six easter eggs are hidden on this site. One of them is a word on this page that keeps learning if you let it.