Mohammed Sinan

AI / Data Engineer, Curanova.ai

Dec 2025 - Present

• Own the deep learning workflow for stroke detection on brain MRI: dataset preparation, training, tuning, validation and inference with nnU-Net and a Swin Transformer in PyTorch.
• Engineer the MRI data pipeline: DICOM and NIfTI handling with pydicom and nibabel, preprocessing, and case-integrity validation across imaging sequences.
• Fine-tune MedGemma with Hugging Face Transformers and LoRA to draft structured medical reports for clinician review.
• Optimize YOLO detection models for real-time vision, including annotation workflows, train and validation splits, and hyperparameter tuning.
• Run GPU training and model serving on Google Cloud (Vertex AI, Compute Engine, Cloud SQL), plus on-premise GPU training for privacy-sensitive patient data.
• Deploy models behind FastAPI services in the Hospital Information System and Hospital ERP, working with developers and healthcare professionals.

Stroke Detection from Brain MRI (Medical Imaging)

• Designed an end-to-end stroke detection pipeline integrating MRI preprocessing, nnU-Net lesion segmentation, Swin Transformer classification, and FastAPI deployment for automated inference.
• Validated each stage independently and ran training on cloud and on-premise GPUs, keeping the flow reproducible from raw study to prediction.

Tech: Python, PyTorch, nnU-Net, Swin Transformer, FastAPI, Google Cloud Platform

Medical Report Generation (Vision-Language Model)

• Fine-tuned the MedGemma vision-language model on medical imaging data to draft structured reports from model findings, giving clinicians an editable starting point.
• Applied parameter-efficient fine-tuning with LoRA and iterated on output structure to keep generated reports template-consistent.

Tech: MedGemma, Hugging Face Transformers, PEFT/LoRA, PyTorch

AI-Powered Hospital Information System (Healthcare AI)

• Integrated an AI chatbot and AI transcription service into clinical workflows, streamlining documentation and patient record lookup for hospital staff.
• Exposed each AI capability as a REST endpoint over PostgreSQL, keeping model services decoupled from core HIS and ERP modules.

Tech: FastAPI, Python, PostgreSQL, React

Fuel Drive-Off Detection (Computer Vision)

• Built a YOLO detection system for fuel drive-off events, assembling and annotating a custom surveillance-image dataset for training.
• Tuned annotation quality, class balance, and detection thresholds across training runs to improve reliability on live camera feeds.

Tech: YOLO, Python, PyTorch, OpenCV

B.Tech, Computer Science and Engineering

MES College of Engineering, Kuttippuram (KTU), 2021 - 2025. CGPA 7.0 / 10.0. Affiliated to APJ Abdul Kalam Technological University (KTU).