ML Researcher — LLM reliability, uncertainty quantification, sequential inference
Builder–researcher focused on reliability, calibration, and robustness in large language models. I design sequential inference systems that provide statistical guarantees on model outputs—work motivated by ensuring AI systems remain trustworthy as they scale.
A sequential LLM inference system that adaptively allocates compute based on confidence. Demonstrates calibrated uncertainty estimates, statistical early stopping, and methods for knowing when models should abstain. Evaluated across MMLU, ARC-Challenge, HellaSwag, and CommonsenseQA.
A scalable data filtering pipeline over 11T tokens, showing that data quality dominates scale. Consistent multi-point MMLU gains on smaller, cleaner corpora.
Reliability checks and validation systems for production ML pipelines. Built automated testing frameworks reducing manual verification by 99%. Working on fault-tolerant distributed systems, observability, and CI/CD.
Built a 135M-parameter Transformer reranker competitive with multi-billion-parameter baselines. Ran controlled ablations on scaling, regularization, and representation quality. Contributed to work awarded SIGIR 2025 Spotlight; first-author publications in ACL Findings and SIGIRD.
Trained domain-specific LLMs using selective pretraining and noise-filtered corpora. HPC-scale training workflows and parameter-efficient tuning.
Advanced NLP, deep learning, computational neuroscience.
National Merit Finalist · James Scholar · ISUR Scholar.
Always interested in conversations about LLM reliability, calibration, and uncertainty quantification — research collaborations, new opportunities, or just a good thread.