Becoming Full-Stack AI Researchers

A fortnightly working group covering the entire LLM stack (training, post-training, serving, agents, evaluation, data infra, interpretability, & safety) with access to an experimental GPU cluster hosted in The Matrix at the Wu Tsai Institute.

Syllabus GitHub Repo

Timeline & Outputs

Fortnightly meetings (short talks, live demos).
Onboarding tutorials for newcomers to LLMs.
Shared GitHub Repository with MWEs, slides, and notes.
Builder Track contributors listed as co-authors on the tutorial paper.

Builder Track

Contribute MWEs, tutorials, and a brief “beyond-basics” showcase. Present at least once. PRs to the repo.

Serve as the local experts on selected topics.
Prepare short demo talks covering fundamentals and advanced features
Contribute to the GitHub Repo
Co-authorship on group write-up

Explorer Track

Attend, learn, and run AI frameworks. Great for interview prep and practical fluency.

Run MWEs yourself
Industry-oriented skills
Optional lightning talks

Tentative Schedule

Location: Room 1167, 100 College; Time: 19:00 (Dinner Starts at 18:30 at WTI Level 11 Lounge);

23 Sept

Inference & APIs, Tools, MCP, Prompt Engineering – Alexander Mader
Distributed Training (Ray, PyTorch vs JAX vs TensorFlow) – George Typaldos & Sasha Cui

7 Oct

SFT (PEFT, Lightning) – Donglu Bai Pretraining - Josh Kazdan

21 Oct

Serving (vLLM) & Distributed Training (DeepSpeed) – Xinyang Hu
MCP (FastAPI) – Alexander Mader

4 Nov

Robotics (OpenVLA, RoboVerse) – Jeffrey Wei, Austin Feng, Oliver Lin
Model Evaluation & Benchmarking (lm-eval, inspect-ai, Terminal Bench) – Chris Zhu

18 Nov

Agents (LangChain, ReAct workflows) – Shivkumar Vishnempet & Xinyang Hu
Interpretability (nnSight) – Oliver Lin, & Will Sanok Dufallo

2 Dec

RL (VERL, Q-function Monte Carlo), Containers (Docker, Apptainer) – Xingzhi Sun, Quan Le, Donglu Bai

Becoming Full-Stack AI Researchers

Timeline & Outputs

Builder Track

Explorer Track

Tentative Schedule

Sponsors

Prof. John Lafferty

Prof. Jas Sekhon

Ping Luo