Emergent cooperation in a pandemic
Can selfish multi-agent RL agents learn to share scarce medical resources? An impact reward cuts deaths 4–11% versus selfish baselines.
Research
Chain-of-thought faithfulness audit
Is chain-of-thought causal, or post-hoc rationalization? Logit lens and activation patching on Llama-3-8B.
Showjumping from broadcast video
Predicting showjumping outcomes with self-supervised video representations, controlling for the venue confound.
Offline RL for menstrual health
Fitted Q-iteration and behavior cloning for personalized recommendations from wearable data.
Learned A* heuristics
Can Q-learning learn better A* heuristics for imperfect mazes with loops?
Products
Civitas
Helps small government contractors find compatible RFPs across 60+ California procurement sites using semantic matching.
Rooted
An AI decision partner and decision history for startup product managers. In progress with Team Geckos, Stanford CS 177.
SAIPIENS
An AI-enabled homework system that turns curriculum PDFs into learning objectives and student units. Top contributor. Stanford CS 224G.
BookClubB
An iOS app for book clubs with groups, discussion threads, and profiles.
ReelLife
A narrative Unity game with daily schedules, character stats, branching dialogue, and multiple endings. Team Reithrodon, Stanford CS 247G.
Contact
I also contribute to crewAI, the multi-agent orchestration framework.