SARTHAK KAR

Dedicated professional skillful in sample collection, data recording and instrument calibration. Resourceful and adaptable individual with extensive experience developing and conducting experiments to document results for studies. Comfortable handling complex issues, meeting strict deadlines and adjusting to rapidly changing conditions.

• Food Vision, 01/2024, 03/2024, Pioneered a Computer Vision project for food category classification in images, employing TensorFlow and the EfficientNet B0 model., Orchestrated data preprocessing using pandas, ensuring uniform image dimensions and scaling within the range of 0 to 1., Enhanced the EfficientNet B0 model's performance by fine-tuning its last five layers, achieving a validation accuracy of 60% with a fraction of the training data. Developed multiple prototypes with various layers of fine-tuning., Currently leading efforts to containerize the model using Docker for seamless deployment.
• SkimLit, 02/2024, 04/2024, Developed SkimLit, an NLP model for abstract sentence classification, leveraging deep learning techniques and various embeddings to enhance accuracy., Created a multimodal architecture capable of handling diverse data types simultaneously, resulting in improved model performance and versatility., Successfully applied SkimLit to make predictions on PubMed abstracts, demonstrating its practical utility for researchers in literature review tasks., Currently working on implementing the model in AWS SageMaker for scalable deployment.
• Movie Prediction, 01/2024, 05/2024, Engineered a movie recommendation system using PyTorch and a dataset of one million movies., Implemented logic to predict movie genres and utilized unexplored data for user recommendations., Transformed user-movie rating data into array format, employing encoding techniques to generate lower-dimensional representations, and decoded them to reproduce the original input vectors., Evaluated data reconstruction fidelity by computing the difference between replicated and original input vectors, ensuring decoding accuracy.
• Heart Disease Analytics, 09/2023, 12/2023, Developed and implemented advanced AI models for heart disease prediction, working closely with data science and engineering teams., Utilized Python programming skills to manipulate, analyze, and visualize heart disease datasets, ensuring data quality and accuracy., Integrated heart disease datasets into common database for centralized data management and analysis, leveraging its powerful data exploration tools., Optimized heart disease analytics pipelines within Snowflake data warehouse, ensuring efficient data processing and query performance for real-time insight.
• Comparison of Diabetes Detection Techniques in Females, 09/2019, 12/2019, Evaluated diabetes detection methods in females with an 80% accuracy rate using machine learning algorithms, including decision trees, Naive Bayes, Logistic Regression, and Neural Network., Adopted Logistic Regression, which achieved approximately 80% accuracy, as the preferred method., Research Paper link given here.