Applied Scientist · NLP & LLMs
Carlos Aguirre
Applied Scientist II at Amazon · PhD, Johns Hopkins · Salvadoreño
LLM Post-training
RL & Preference Optimization
Agentic AI
Fairness & Trustworthy NLP
Applied Scientist · NLP & LLMs
Applied Scientist II at Amazon · PhD, Johns Hopkins · Salvadoreño
I'm an Applied Scientist II at Amazon, where I work on post-training and alignment of large language models for conversational AI — reinforcement learning and preference optimization, reward modeling, and LLM-based evaluation, with a growing focus on agentic AI and the environments used to train and evaluate it.
I completed my PhD at Johns Hopkins University, affiliated with the Center for Language and Speech Processing and advised by Mark Dredze. My research centered on measuring and improving the fairness of NLP models — including methods for training under limited demographic information, applications in mental-health NLP, and human-AI interaction. Trustworthy, human-centered AI still shapes how I approach applied work today. Outside of research, I teach traditional bachata.
From production LLM releases to research on fairness and agents.
Reinforcement learning (PPO/GRPO-style methods) and preference optimization (DPO), reward modeling, and multi-objective training for production conversational assistants.
Tool-using agents, agentic benchmarks, and the training/evaluation environments ("gyms") behind them — plus LLM-as-a-judge evaluation and the infrastructure to run it at scale.
My doctoral research: measuring and improving fairness in language models, learning under limited demographic signal, and responsible NLP in sensitive domains like mental health.
Post-training and alignment of large language models for conversational AI: RL & preference optimization, reward modeling, LLM evaluation, and agentic AI. Promoted from Applied Scientist in 2026.
Affiliated with CLSP, advised by Mark Dredze. Research on fairness of NLP models, learning under limited demographic information, and mental-health NLP.
Undergraduate research assistant in the KDD Lab with William Hsu.
Grew up in El Salvador (Colegio Salesiano Santa Cecilia) before moving to Kansas City.
Off the keyboard, I'm on the dance floor.
I teach traditional (Dominican) bachata at Salsa con Todo — sharing the roots, musicality, and connection of the dance with a growing community.
Salsa con TodoDance keeps me grounded and connected to my Latin American roots. Teaching traditional bachata is my way of passing on a tradition that's equal parts rhythm, history, and human connection — the same curiosity about people that drives my work in human-centered AI.