Soumik Hati

Data Science and AI professional with hands-on experience delivering end-to-end Machine Learning and Deep Learning solutions in NLP and Computer Vision. Developed a scalable Extractive Text Summarization system using Python and TF-IDF with multi-format input support, focused on efficiency and real-world usability. Built a healthcare AI pipeline for histopathological image analysis using U-Net for tumor segmentation and ResNet/VGG for classification, achieving 92%+ accuracy with strong Dice and IoU metrics. Strong in data preprocessing, model evaluation, and performance optimization. Passionate about building scalable, business-impacting AI solutions and contributing to high-performance teams in global organizations.

Histopathological Image Segmentation and Classification (June 2024-Dec 2024): Built an end-to-end Deep Learning pipeline for automated histopathological image analysis to support accurate cancer diagnosis. Implemented U-Net for tumor and nuclei segmentation, achieving strong Dice and IoU scores. Developed CNN-based classification models using ResNet and VGG to classify tissue samples (benign vs. malignant), achieving 92%+ accuracy on BreakHis and Camelyon16 datasets.
Applied preprocessing techniques including color normalization, patch extraction, and data augmentation to improve model generalization. Performed model evaluation using accuracy, Dice coefficient, and IoU metrics.
Utilized Python, Deep Learning, Computer Vision, CNN architectures, and medical image processing to deliver a scalable healthcare AI solution.

Extractive Text Summarization Tool (NLP, TF-IDF) (Jan 2025-Jun 2025): Developed an end-to-end Extractive Text Summarization system using Python and NLP to generate concise summaries from large documents. Implemented a TF-IDF–based sentence ranking algorithm to identify and extract the most relevant information with high accuracy and readability.
The application supports multi-format inputs including direct text, TXT/PDF uploads, and Wikipedia links, ensuring flexibility and real-world usability. Designed the solution to be lightweight, computationally efficient, and scalable.
Demonstrated expertise in Python, Natural Language Processing, text preprocessing, feature extraction, and algorithm optimization while delivering a practical AI-driven solution.