Krish Desai

Krish Desai

Parkinsons Disease (PD) is a neurodegenerative disorder resulting in motor deficits due to advancing degeneration of dopaminergic neurons. PD patients report experiencing tremor, rigidity, visual impairment, bradykinesia, and several cognitive deficits. Although Electroencephalography (EEG) indicates abnormalities in PD patients, one major challenge is the lack of a consistent, accurate, and systemic biomarker for PD in order to closely monitor the disease with therapeutic treatments and medication. In this study, we collected Electroencephalographic data from 15 PD patients and 16 Healthy Controls (HC). We first preprocessed every EEG signal using several techniques and extracted relevant features using many feature extraction algorithms. Afterwards, we applied several machine learning algorithms to classify PD versus HC. We found the most significant metrics to be achieved by the Random Forest ensemble learning approach, with an accuracy, precision, recall, F1 score, and AUC of 97.5%, 100%, 95%, 0.967, and 0.975, respectively. The results of this study show promise for exposing PD abnormalities using EEG during clinical diagnosis, and automating this process using signal processing techniques and ML algorithms to evaluate the difference between healthy individuals and PD patients.

Emi Soroka, Tanmay Chopra, Krish Desai et al.

Large language models (LLMs) have seen increasing popularity in enterprise applications where AI agents and humans engage in objective-driven interactions. However, these systems are difficult to evaluate: data may be complex and unlabeled; human annotation is often impractical at scale; custom metrics can monitor for specific errors, but not previously-undetected ones; and LLM judges can produce unreliable results. We introduce the first set of unsupervised metrics for objective-driven interactions, leveraging statistical properties of unlabeled interaction data and using fine-tuned LLMs to adapt to distributional shifts. We develop metrics for labeling user goals, measuring goal completion, and quantifying LLM uncertainty without grounding evaluations in human-generated ideal responses. Our approach is validated on open-domain and task-specific interaction data.