Alexander Waitz

I recently graduated from Stanford, where I earned my MS/BS in Computer Science.

I was extremely fortunate to have been advised by Simran Arora at Hazy Research.

My research interests include:

• Model memory. How should models retain and reason with information over the course of hours, weeks, months, years?
• Hardware-software co-design. How can algorithmic primitives, like sequence mixers, be redesigned to efficiently map to hardware?

Things I've done, in decreasing order of relevance:

• Trained audio models at Cartesia
• Designed memory-efficient LLMs with Scaling Intelligence Labs
• Worked on kernels at d-Matrix
• Built profiling tools for Legion
• Studied complexity theory and ML at Oxford
• Biked from San Francisco to Washington D.C. with Stanford Spokes
• Designed and taught "The Art of Walking"
• DJ'ed at KZSU

Projects:

• [New work coming soon! Under review...]
• Test-Time Training for Efficient RL Sequence Modeling
  • TTT for efficient RL.
• Fast Inference with Dynamic Tree Speculative Decoding
  • Entropy-driven approach for faster speculative decoding.
• Megatron
  • Agentic coding tool.
• TestNinja
  • Context-aware test generation.
  • 1st place & "Most Interesting Technical Achievement" at South Park Commons x Meta AI hackathon.
• Gaussian Closet
  • Diffusion in-fill for clothing try-on.
• Ray Tracer
  • Rust ray tracer compiled to WASM.

Contact:

• My email is last name at cs.stanford.edu
• Linkedin
• Github