Shital Shah · Research Engineer, Microsoft Research

If universe is an optimizer, what is its loss function?

Research Engineer at Microsoft Research with interests in deep learning and reinforcement learning. Created AirSim, TensorWatch, archai, most recently worked on Phi series of models.

Welcome! 🎉

In this world of the Greater Being, I feel like I’m at play… like a little kid. With all my sense of wonder and curiosity… walking on my knees to get close, see and feel everything that has been put around me… and smile for all those good things that I find.

Check out my blog posts.

For more latest and greatest follow me on Twitter.

You can find a lot of my hobby projects on GitHub.

You can find research papers I contributed to at Google Scholar.

Recent 916 posts in the chain

  1. Releasing StreamHope - Merge Duplicate Playlists in YouTube Music, transfer from Spotify

    2 min read

    Since past week, I have been finishing off several of side projects that were on my mind. The thing is that I haven’t written any code for these projects, they are all 100% done by AI (in this case Claude Code/Opus 4.5).…

  2. How to Get and Interpret GPU Memory Profiling

    6 min read

    The torch.OutOfMemoryError is one of the most dreaded errors where many folks descend into making guesses until something works. A much better, faster, and disciplined approach is to profile memory utilization. Like…

  3. o3-mini Level Hard Problem

    tweet
  4. DeepSeek's Recipe Evolution

    tweet
  5. Can You Spot Error Proving 1 = -1?

    tweet
  6. Phi-4 Now On HuggingFace with MIT Licence!

    thread
  7. Phi-4: The Most Powerful smol Model!

    thread
  8. Announcement: PhD Internship for AI for Reasoning

    thread

all posts →