PhD student in Electrical Engineering at Stanford, co-advised by Dr. Emily Fox and Dr. John M. Cioffi. I work on reinforcement learning for verifiable rewards, preference optimization, test-time scaling, and agentic systems for reliable reasoning models.

Collaborations across Google DeepMind, Meta MSL, Amazon AGI, and Microsoft Core AI. Building alexein.ai, an agentic platform for collaborative AI agents.

Preference Optimization & Test-Time Scaling

Sample-efficient preference-learning and RL methods for aligning language-model reasoning, plus adaptive test-time compute that controls inference depth, tool invocation, and verification under strict budgets. I study how post-training and inference-time search jointly shape reasoning, and the trade-offs among accuracy, latency, and reliability.

General Preference RL Continuous-Utility DPO Adaptive compute Coverage-preserving RL

Agentic System Architectures & Coding Agents

Agentic systems that plan, act, and self-verify over long horizons, spanning multi-agent coordination and the software agents I develop at alexein.ai. At Amazon AGI I study how coding agents convert test-time compute into solved tasks by exploring diverse solutions and writing trustworthy verifiers, and how those behaviors are internalized through supervised and reinforcement post-training.

Teaching Agents to Code Reliably SWE-bench Verified GRPO self-verifiers

AI Alignment & Benchmarking

Persona-based benchmarks and user-simulation frameworks that evaluate how models behave across diverse populations and interactive settings. I care about calibrated uncertainty, reducing sycophancy, and measurement methodology that makes alignment progress observable and reproducible.

PersonaEval MatrAIx Sycophancy disentanglement Reward decomposition

Fellowships & Medals

• Stanford Graduate Fellowship
• Qualcomm Fellowship
• Knight-Hennessy Fellowship (Finalist)
• President's Medal · Rector's Gold Medal

Service

• Area Chair — ICASSP 2026, NeurIPS 2025
• Workshop Co-Chair — IEEE ICC 2026
• Reviewer — NeurIPS, ICML, ICLR, AAAI, KDD, CoLM
• Best Paper — IEEE ICC, IEEE FIT 2026