Devam Sheth Email & Phone Number

Bellevue, Wa, Us

Tempe, Az, Us

Phoenix, Arizona, United States

Tempe, Arizona, United States

● Utilized cutting-edge nn-Unet techniques to enhance white matter segmentation in Brain MRI data, achieving a mean dice score of 0.65 across an ensemble of 5 cross-validation folds. Conducted experiments efficiently on ASU GPU clusters.● Specializing in developing Visual Language models that can analyze both image and text data to identify and classify headache types from brain MRI, leveraging and researching models like CLIP, Flamingo,Lit, ConVIRT and BIOMedCLIP.● Preprocessed and managed complex medical image and text datasets, ensuring data quality for robust model training.

● Researched and analyzed innovative approaches for lowlight enhancement in computer vision, with a focus on improving performance while reducing inference time.● Developed a PyTorch and Keras model utilizing advanced loss functions inspired by SOTA lowlight enhancement models, while leveraging the efficient Squeezenet Architecture to optimize inference time and enhance lowlight images while maintaining the PSNR value to 25 for reconstructed images. Reduced model parameters from 79k to 14.3k and MACs from 5.19G to 926.42M.● Integrated fire module inspired by Squeezenet Architecture to further reduce MACs, also conducted hyperparameter search experiments on NVIDIA TITAN V. Performed inferencing on images and videos, obtaining enhanced outputs.

Ahmedabad, Gujarat, India

● Conducted extensive research on detecting and performing multi-class classification of 6 bearing fault types using Image classification from Case Western Reserve University data of bearing vibrations recorded at 2 different frequencies.● Compared various models and preprocessing approaches to identify the best classification method based on performance metrics of vibration data over different Motor RPMs, achieving robust classification results.● Developed a method for bearing fault classification that demonstrates superior generalization with a 98% mean accuracy over a range of Motor RPMs compared to previously published works.

Ahmedabad, Gujarat, India

Got familiarized with the whole process flow of Heat Exchangers and Pressure vessels from procurement to end-product manufacturing, quality, and logistics.Worked with the Design team, helped them to automate the design calculation process by developing a program based on Python using basic OOP concepts also proposed to build a GUI related to the software.

Master Of Science - Ms, Computer Science, 4/4

Activities and Societies: Math Tutor @ America Reads

Bachelor Of Technology - Btech, Mechanical Engineering

Minor Specialization, Computer Science

Senior Secondary (Xii), Science, Cgpa :8.90/10