I build responsible AI systems that support clinicians with medical imaging, screening, and automated reporting.
Sayan K. · Medical AI Specialist · Remote
Focus
Medical imaging · Vision AI · LLM agents
Tools
Values
Clinical responsibility, privacy awareness, transparent metrics.
I design AI systems that assist clinicians rather than replace them. My work focuses on medical imaging, early screening support, and workflow automation, often prototyped on Kaggle GPUs using transfer learning techniques.
I strictly adhere to using public research datasets and privacy-aware practices to ensure ethical development.
Imaging and workflows for healthcare and research.
Bridge between visual models and language agents.
Models used for research and decision support, not stand-alone diagnosis.
Specialized technical services for medical AI development.
Classification and detection models for tasks like early screening, quality control, and research workflows.
Fine-tuning SOTA architectures (ViT, EfficientNet, YOLO) on your dataset, with clear metrics and documentation.
LLM-powered agents that turn model outputs into draft clinical-style reports and triage suggestions.
Lightweight APIs, Dockerized services, and secure, privacy-aware deployment options.
All models and demos are intended to support research and clinical workflows and are not certified medical devices or stand-alone diagnostic tools.
Recent work in medical vision and automation.
A vision model that analyzes palm and fingernail images to assist with early anemia screening. Trained on public research data to demonstrate non-invasive screening feasibility.
Objective: Early screening support, not diagnosis.
Approach: Transfer learning (CNN/ViT) with color normalization.
"Outputs a 'possible anemia' score for clinician review."
Disclaimer: For research/educational purposes only. Not a diagnostic tool.
A YOLO-based prototype for detecting dermatological lesions from dermoscopy images (ISIC dataset), aimed at assisting clinicians with triage workflows.
Plan: Localize suspicious lesions with bounding boxes.
Focus: Interpretability and reducing false negatives.
A pipeline connecting vision model findings with an LLM agent to draft structured, clinician-friendly reports ('Findings' & 'Impression').
Goal: Draft reports for clinician approval.
Privacy: Designed for on-prem/controlled environments.
A structured approach to clinical AI development.
We define the clinical question and understand your data constraints.
Train and evaluate an initial model on a subset of data.
Refine metrics, improve robustness, and review with stakeholders.
Deploy APIs or agents that fit into existing clinical workflows.
"My work focuses on responsible AI that complements clinical expertise, not replaces it."
Interested in collaborating on medical imaging, screening tools, or workflow automation?
Feel free to reach out and share a brief description of your project.
Responses are typically within 1–2 business days.