Profile
I build and evaluate LLM agent systems and multilingual, multimodal AI, with a focus on the infrastructure that makes them reliable: agent architectures, evaluation frameworks, and human-in-the-loop pipelines. My work spans text and speech systems, translating behavioral and cognitive science into practical AI for real-world deployment.
News
-
2026-Sept: New open access paper (to appear in TMLR) introduces the Blicket benchmark, a causal-reasoning test of whether LLM agents can abandon outdated hypotheses and build better ones
-
2026-Sept: New machine learning study and dataset on gender assignment in Hindi (accepted at the 6th Multilingual Representation Learning (MRL) Workshop 2026, co-located with EMNLP, Budapest, Hungary)
-
2026-Aug: New open access article in Medicina finds that a guardrailed, rule-based conversational agent achieves outcome parity with human health coaches on 6-month weight loss, via unsupervised phenotyping of engagement in a 9,470-patient tirzepatide cohort
-
2026-Aug: New article in Diabetes, Obesity and Metabolism examining how AI-driven patient support relates to 6-month medication adherence in a real-world retrospective digital obesity cohort
-
2026-Mar: New computational study of morphological predictability in Tashlhiyt Berber, accepted for SLiDE workshop at LREC (Mallorca Spain, May 2026)
-
2025-May: New slides and conference article on ASR evaluation and a new benchmark dataset in INTERSPEECH 2025
