# Resume: Peter Li

- In the room: <https://peterli.space/#resume>
- Index: <https://peterli.space/llms.txt>

PDF: https://peterli.space/resume.pdf

## Experience

### Research Affiliate, Massachusetts Institute of Technology
04/2026 - present. Trustworthy medical AI. Remote. Working under L. C. and M. K.

- Lead author of Ask Before You Answer, a clinical-LLM calibration framework under review at NeurIPS 2026.
- Built a LoRA training and five-seed evaluation pipeline for MedGemma-27B, raising active clinical inquiry from 17.5% to 45.6% while preserving aggregate answer quality.
- Co-authored Agents Catching Agents, on benchmark gaming in clinical multi-agent systems (2nd Agentic AI for Medicine Workshop @ MICCAI 2026).

### Founder, Substrate Labs
05/2026 - present. Independent AI-safety research lab. West Vancouver. [substrate-labs.org](https://substrate-labs.org)

- Founded and lead an independent AI-safety research lab, recruiting a high-performing team across mathematics, science, industry, and academia.
- Four NeurIPS 2026 workshop acceptances: a solo-author oral (BiAlign), first-author posters (HAIC, SocialAgent), and a Meta-Agents poster.
- Led a second paper submitted to AAAI 2027 on LLM-agent reliability in financial markets.
- Authored formal public comments on frontier-AI governance submitted to the Colorado Attorney General, U.S. FTC, and European Union.

### Member of Technical Staff, Moonshot (YC S26)
06/2026 - 09/2026. Remote.

- Worked on SLMs and other side quests.
- Founders were incredibly interesting (quite contrarian). Very bright future ahead.

### Research Intern, UC Santa Cruz
06/2026 - present. Remote. Working under L. G.

- Research internship in mechanistic interpretability.

### Work Experience, HUB International
05/2026 - 07/2026. Apprenticeship. Vancouver, British Columbia, Canada. On-site.

- Learned from people working at the sixth-largest insurance brokerage in the world.

### Research Assistant, Stanford University
04/2026 - 07/2026. Working under K. S. and P. K.

- Research assistantship.

## Education

- Palantir Technologies: Fellowship
- Y Combinator: Startup School (in person). San Francisco. July 2026.
- BlueDot Impact: Technical AI Safety. July 2026.

## Skills

- Machine Learning: LoRA / supervised fine-tuning, sparse autoencoders, linear probes, activation steering, evaluation-pipeline design
- Tools: PyTorch, Hugging Face Transformers, PEFT
- Focus: epistemic calibration & abstention, mechanistic interpretability, chain-of-thought faithfulness, evaluation awareness
- Certifications: CMSA® (CFI), Data Analytics Essentials (Cisco), Enterprise Design Thinking Practitioner (IBM)
