ML pipelines for intent classification, sentiment analysis, and routing prediction on customer transcripts. Integrated RAG and LLM-based task classification.

Multi-trigger classification in mechanistic interpretability under Dr. Jan-Philipp Fränken. Achieved ~98% AUROC on single-classification accuracy.

First-author EMNLP 2025 on LLM benchmarking, first-author ACL 2023 on symbolic math reasoning. Built evaluation protocols for symbolic reasoning datasets.

Led LLM development for IKEA dashboard navigation. Modeled marketing spend for Zillow. Developed clustering algorithms for Fox viewer personas.

Led clinical testing for Transformer-based (F1 ~0.87) tool for detecting stutters in speech, working with 20+ SLPs and academics.