Stefan Bhagwandin
AI Tutor Image And Video at xAI
Old Toronto, Ontario, Canada
Summary
🤩
Rockstar🎓
Top SchoolStefan Bhagwandin is a multidisciplinary technologist and storyteller with 11 years of experience blending game development, technical writing, and AI data work. He shipped a 120k-word indie adventure-platformer, Monospaced Lovers, across PC, PS5, and Switch and previously reached 200k downloads with a free visual novel, demonstrating rare end-to-end product chops from code (Unity/C#) to art, narrative, and launch marketing. More recently he’s contributed to improving video and audio understanding for Grok at xAI, doing QA, labeling, and preference evaluation while onboarding peers. On the engineering side he has hands-on ML optimization experience—authoring custom Triton/CUDA kernels and FP8 quantized GEMM work for a fast-serving LLM/VLM framework. Comfortable switching between creative and technical domains, he favors clear, colorful prose and pragmatic tooling over buzzwords.
11 years of coding experience
7 years of employment as a software developer
Honours Bachelor of Arts Film Studies (Major) & English Philosophy (Minors), Honours Bachelor of Arts Film Studies (Major) & English Philosophy (Minors) at University of Toronto - Victoria UniversityMary Ward Catholic Secondary SchoolInternational Diploma Marketing, International Diploma Marketing at UC Berkeley Extension
Github Skills (14)
cuda10
matrix-multiplication10
pytorch10
p810
triton10
deep-learning10
inference10
cprogramming-language9
benchmark9
benchmarking9
c-language9
mlops8
llm7
transformer7
Programming languages (6)
Github contributions (5)
sgl-project/sglang
Jan 2025 - Apr 2025
SGLang is a fast serving framework for large language models and vision language models.
Role in this project:
ML Engineer
ML EngineerContributions:43 reviews, 26 PRs, 4 pushes in 2 months
Contributions summary:Stefan's contributions center around optimizing and benchmarking FP8 (8-bit floating point) matrix multiplication kernels for the SGLang framework. They implemented a custom Triton kernel for grouped GEMM operations, optimized the existing kernel, and integrated a new per-token quantization function using CUDA. Further work included adding support for W8A8 quantization and ensuring the correct functionality for online quantization, as well as testing and benchmarking the kernels.
cudadeepseekdeepseek-llmdeepseek-r1deepseek-r1-zero
hebiao064/sglang
Feb 2025 - Apr 2025
SGLang is a fast serving framework for large language models and vision language models.
Contributions:13 PRs, 144 pushes, 23 branches in 1 month
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.