Training and evaluation for voice agents
Built an RL environment and training pipeline for instruction-following voice agents, then developed synthetic-caller evaluations to stress-test production systems across adversarial scenarios.
ML researcher & engineer
Building systems that make machine learning useful in the real world.
I studied Computer Science with a Machine Learning concentration at Carnegie Mellon. My work spans training and evaluation for conversational agents, model distillation, and research on how neural networks learn and forget.
Selected work
Research and systems
Diffusion policy for robotic imitation learning
Implemented and evaluated a transformer-based diffusion policy that generates multi-step robot actions by denoising trajectories conditioned on state history.
Learning with diffusion models
Built transformer-based speech recognition and diffusion models for multi-step financial forecasting, combining market data with sentiment embeddings from more than 80,000 tweets.
Experience
Applied ML, research, and teaching
Stealth startup
Founding Engineer
Developed RL training, prompt optimization, and adversarial testing infrastructure for conversational voice agents.
Moloco
Machine Learning Engineering Intern
Worked on revenue-model distillation and data pipelines; automated the workflow with Airflow and built reusable training-data augmentation in GCP Dataflow.
Apple
Machine Learning Engineering Intern
Improved performance-anomaly detection, built correlation tooling for engineers, and trained an autoencoder on Vision Pro performance traces.
Carnegie Mellon University
Researcher · Teaching Assistant
Researching fine-tuning, overparameterization, and catastrophic forgetting; teaching Mathematical & Computational Foundations for ML.
MIT PRIMES
Researcher
Advanced a more fault-tolerant dynamic consensus protocol and presented the work to researchers at MIT.
Elsewhere