Varun Shitole
Angestellt, Data Scientist, Spryfox GmbH
Bis 2022, Computational modelling and Simulation - Visual Computing (ML & CV), Technische Universität Dresden
Darmstadt, Deutschland
Über mich
I have been working as a Data Scientist since Aug 2022 and have a strong background in applied machine learning and data science across various domains. Additionally, I have gained meaningful experience in client handling, project management, and interdisciplinary team communication. I am looking for an exciting opportunity to learn and grow further in these fields.
Werdegang
Berufserfahrung von Varun Shitole
Implementing and validating ML algorithms for various use cases like healthcare, Image processing, and NLP. Successfully delivered an interactive and automated Power BI dashboard to view key statistics. Ensured quality EHR data by developing quality metrics and working with respective stakeholders. Based on data insights, Standard operating procedures were introduced to tackle the problems at the source Detected Names, medicines, and relevant clinical information from free text documents using Spark NLP
- Self-supervised learning is based contrastive approach for scientific Image datasets. Focus on Investigating how contrastive methods can be tailored for scientific datasets like medical imaging, microscopy, etc. - Systematically evaluated Image Features/Embeddings generated with different cropping strategies - Evaluated the effect of transfer learning in low data regime - Explored contrastive Methods like SimCLR, Barlow twins, DINO, and SwAV. - Implementing Data processing and Integration Pipelines
Ausbildung von Varun Shitole
5 Monate, Nov. 2020 - März 2021
Research Project - Physics-Informed Neural Networks for Materials Science
TU Dresden
- Developed a surrogate for DFT calculations – called PINN-DFT – that is computationally more efficient by orders of magnitudes while maintaining the accuracy of DFT. - Deep learning framework to solve PDEs using prior knowledge about the problem. The problem statement solved the Kohn-Sham equation, time-independent form. In the future, It will enable to overcome the computational bottleneck of DFT and enable electronic structure calculations for much larger systems.
4 Monate, Apr. 2020 - Juli 2020
Project - Segmentation and classification of objects in virtual microscopy slides.
TU Dresden
- Masked segmentation approach with subsequent multi-classification on simulated data to maximize performance for a dataset with multiple classes - U-Net architecture used for segmentation
2 Jahre und 8 Monate, Okt. 2019 - Mai 2022
Computational modelling and Simulation - Visual Computing (ML & CV)
Technische Universität Dresden
Machine Learning Statistics Relational Database Visualization Computer Vision
4 Jahre, Juni 2012 - Mai 2016
Mechanical Engineering
University of Pune
Sprachen
Deutsch
Gut
Englisch
Fließend
Hindi
-