Vaibhav Rao

Software Development Engineer with over 7 years of experience in Software development, focusing on AI and NLP. Successfully enhanced conversational AI interfaces and implemented cryptographic security measures across cloud platforms, resulting in a significant increase in system efficiency. Proven proficiency in advanced programming languages and machine learning models, contributing to robust solutions in high-stakes environments. Committed to leveraging innovative technical strategies to propel advancements in software development.

• Developed linear classification models using the Stochastic Gradient Descent classifier to classify FNA cells as benign or malignant on the Wisconsin Diagnostic Breast Cancer dataset and a deep learning model using CNN architecture for predicting classes with 99% accuracy on the German traffic signs dataset with 70,000 sample images using Keras/TensorFlow (Python).
• Developed an OCR tool to recognize scale and font variant images to identify characters implementing the connected component labeling algorithm implemented depth first search using some modules from OpenCV (SIFT, Homography, RANSAC).
• Implemented computer vision projects using OpenCV for combining images in a panorama, face clustering using Face-cascade and K-Means clustering, detecting celebrity faces, and classifying them.
• Implemented a classification algorithm for MNIST dataset using 0-1 loss function Bayesian Decision Rule without standard library modules. Compared the results with better algorithms like K-NN, SVM's and Neural Networks to understand drawbacks. Built graphs on R/Tableau.
• Used ROS operating system to implement algorithms on F1Tenth Platform for obstacle avoidance, robot control, path planning.
• Developed various robotics algorithms on ROS using LIDAR, A-star path planning, visual/depth cameras, visual odometry, mapping, SLAM and Bayesian Estimation, including creating an enhanced RESNET CNN architecture and a semantic segmentation pipeline using UNET.
• Developed a Nano PyTorch module implementing all CNN Backpropagation layers using plain NumPy understanding the gradient flow, creating the Neural Network graph and implementing layers like Dense, Convolution, Pooling and Loss Functions.
• Develop an enhanced RESNET CNN architecture to classify TinyImage/Cars196 dataset. Enhancing accuracies from RESNET18. Also developed a Semantic Segmentation pipeline using UNET to perform facial attribute segmentation in Computer Vision with custom Data Loader achieving up to 65% accuracy on the standard dataset.
• Implemented Q-Learning/SARSA/TD algorithms in RL to solve grid environments, developed Deep Q-Networks to solve OpenAI gym environments, and implemented deep convolution neural networks for ATARI games.
• Developed and worked with various NLP models like Transformer/BERT/ELMo as part of my course on Computational Linguistics.
• Created a GAN-based image colorization model for self-supervised learning of visual image features and coloring black and white images.
• Won University at Buffalo Hackathon 2021 implementing a CNN Based Model to detect facemasks and an Automated dispenser on Arduino to dispense masks to non-masked people. https://devpost.com/software/frontline-ai