Projects
Selected Projects
SAFE-ICU Data Resource
SAFE-ICU is one of the most important outcomes of my PhD work. It is a freely available de-identified pediatric intensive care data resource developed from patients admitted to the PICU at AIIMS New Delhi, and it supported a broad set of machine learning and translational AI studies in critical care.
Beyond being a dataset, SAFE-ICU grew into a larger research ecosystem that included predictive modeling, multimodal monitoring, clinical dashboards, and deployable AI tools. This work shaped much of my PhD research and remains one of the central achievements in my profile.
- Project page: SAFE-ICU
- Previous lab page: TavLab profile
- Google Scholar: Publications
SIgnose
SIgnose is a relevant ICU-focused machine learning project in the TavLab ecosystem and a better representative software artifact for my website than older repositories I no longer want to feature.
- Paper: SSRN abstract 4248013
- GitHub: tavlab-iiitd/SIgnose
ThermalShockNet
ThermalShockNet is my work on thermal imaging-based shock prediction using deep learning on thermal videos. This line of work focuses on non-contact monitoring and early warning for hemodynamic shock in pediatric intensive care settings.
- Paper: Early Prediction of Hemodynamic Shock in Pediatric Intensive Care Units With Deep Learning on Thermal Videos
- GitHub: Shock-Prediction-with-Thermal-videos
ContraIndicator
ContraIndicator is an integrated tool designed to detect, visualize, and help prevent potential drug–drug interactions in clinical practice, especially pediatric ICU workflows. It is directly tied to the pediatric ICU DDI research line and is a strong example of turning research into a usable software tool.
- Paper: Springer article
- GitHub: tavlab-iiitd/ContraIndicator
ThermoGnose
ThermoGnose is the pipeline I developed for early prediction of hypothermia in pediatric intensive care using routinely collected physiological vitals. The work used SAFE-ICU data and focused on clinically meaningful early-warning horizons.
Generalizing ICU AI
This medRxiv preprint studies how intensive care AI can generalize across time scales in resource-limited settings and fits well with my broader work on ICU AI and translational machine learning.
FAIR Healthcare Data Harmonization
Another major part of my work has focused on making healthcare datasets more useful for downstream analysis and machine learning. This included harmonizing non-curated health data using NLP and LLM-supported pipelines to improve interoperability and analytical readiness.
Foundation Models for Biology Data
My current work at INRIA focuses more strongly on computer science and machine learning methods for biology data, especially small foundation models, representation learning, and practical model design for scientific use cases.