PhD Researcher, IIT (BHU) Varanasi
Co-Founder at 
Hridae
I work across neuroscience, AI, biosignals, and medical images to build clinical decision support tools, agentic harness and wearable hardware.
01About
My PhD develops EEG-based screening and triage tools and systems that support clinical decisions in neurology, with a focus on dementia and epilepsy. I aim for tools that are interpretable, so a clinician can see which features of the signal led to a result, and that keep working on patients from hospitals not used to build them.
Beyond EEG, I build automated processing pipelines for different medical imaging modalities, and a lightweight local agentic harness and orchestration layer on open models, for long and complex tasks. At Hridae Medical Technologies, which I co-founded, we develop wearable monitoring, ECG models and ML-accelerated digital twins from medical images for precision cardiology and neurology. I studied physiology and neuroscience before engineering. I collaborate with clinicians and researchers at the Institute of Medical Sciences (BHU), AIIMS New Delhi, IIT Kanpur and IIT Kharagpur.
Soinik Ghosh is a PhD researcher at the School of Biomedical Engineering, IIT (BHU) Varanasi, and a co-founder of Hridae Medical Technologies. The PhD focuses on EEG-based screening, triage and clinical decision-support tools for dementia and epilepsy. Work on reduced-electrode EEG for cognitive screening, carried out with neurologists at the Institute of Medical Sciences, BHU, was accepted at the GlobalSouthAI workshop at NeurIPS 2026. Soinik also builds automated processing pipelines for medical images and a lightweight local agentic harness on open models. Hridae develops wearable monitoring, ECG models and ML-accelerated digital twins from medical images for precision cardiology and neurology. Collaborators include clinicians and researchers at IMS BHU, AIIMS New Delhi, IIT Kanpur and IIT Kharagpur. Soinik studied physiology and neuroscience at the University of Calcutta before moving into engineering.
02Featured
GlobalSouthAI @ NeurIPS 2026 · Accepted
In recordings re-derived to simulate low-channel devices, a few well-placed electrodes screened for cognitive impairment about as well as a full clinical EEG. This points towards simpler screening devices, which still need testing as real hardware.
03Research
Tools that help clinicians decide who needs a full cognitive assessment, using simpler EEG recordings.
Methods that adapt to each patient's EEG to detect seizures and estimate when seizure risk is rising.
Pipelines that prepare, check and analyse images from different modalities with few manual steps, so imaging studies run the same way each time.
A harness and orchestration layer on open models, for long and complex tasks. It runs on an 8 GB laptop GPU, so data stay on the device.
Wearable monitoring devices, ECG models, ML-accelerated digital twins from medical images and analysis software, developed with clinicians.
hridae.com
Short cognitive tasks that run offline on a phone or laptop and track changes in each person over time. Designed to support screening, not to diagnose.
04Work
Generates patient-specific synthetic EEG to train seizure detectors when patient data are limited.
A low-cost single-channel EEG device that processes signals in real time.
A prototype for AI-assisted, non-invasive heart disease screening.
A virtual-reality guided robotic exosuit for stroke rehabilitation.
Explainable medical image analysis that runs on small, low-power devices.
05Publications
06Background
Wearable monitoring for precision cardiology and neurology.
IIT (BHU) Varanasi
EEG-based screening, triage and clinical decision-support tools.
In collaboration with IMS BHU, AIIMS New Delhi, IIT Kanpur and IIT Kharagpur.
University of Calcutta
Surendranath College, University of Calcutta
07Art
Open to research and engineering roles, internships and collaborations in clinical AI, medical imaging, on-device AI and AI agents.