Projects
Trace2Evolve: A Karpathy-Style AutoResearch Harness for Customer-Support Agents
May 2026
A research-style AutoResearch harness for improving tool-using customer-support agents using benchmark traces, heldout tau2 evaluations, synthetic pressure tests, and reliability-gated promotion.
Teaching an LLM to Explore: Reinforcement Learning for Document Navigation
March 2026
PDF | Code | Read more →
Privacy Guard: Learning Privacy-Budgeted Active Sensing Policies via Reinforcement Learning in Smart Home Environments
March 2026
PDF | Code | Read more →
Training a Deep Q-Network to Master Uno: A Comprehensive Study in Reinforcement Learning for Imperfect Information Games
January 2026