Projects

Research under Prof. Beth Anne Helgason at the Yale School of Management on paltering in AI models: cases where a model doesn't state anything false, but phrases its answer so the user comes away believing something that isn't true. Working on ways to detect it.

• AI safety
• LLM evaluation
• Deception

Developer on Yale Meals, an open-source app from the Yale Computer Society for viewing dining hall menus and other meal-related statistics. A large share of Yale students use it.

• Open source
• Mobile
• Yale Computer Society

Early-stage work with Prof. Arman Cohan on how well LLMs reason across multiple research papers at once.

• LLMs
• Reasoning
• NLP

Refined a transformer-based deepfake audio detection model at StepFun, pruning and retraining it to run inside the production data pipeline. Leaned on interpretability methods like whitening the representations, early-exit tests to find the depth at which the model had already decided, and AUC probing layer by layer.

• Audio ML
• Interpretability
• Model pruning

Studied how AI agents influence human self-perception by building a multi-agent simulation in which LLM agents and human participants engaged in self-revelatory dialogue. Found measurable self-view convergence in social contexts, but not in perceptual ones. Project lead.

• LLM agents
• Human subjects
• Python

Applied Louvain community detection and k-means clustering to model how social organizations evolve, focusing on predictive analytics in social science contexts. Completed through the Pioneer Research Program.

• Network analysis
• Clustering
• Python

Led a team designing and evaluating a mobile mini program for medication management in chronic kidney disease patients. The paper was accepted into the CTB database, and the project advanced to the global finals and took third place.

• Mobile
• Health
• Team lead