I'm a final-year student, studying Electronics and Communication at Indian Institute of Technology, BHU Varansi, India. I have always been fascinated by the recent advances in the field of Artifical Intelligence from Deep Learning to Deep Reinforcement Learning. I mainly study and work in the field of computer vision and deep reinforcement learning and occasionally on natural language processing. I enjoy solving problems and actively take part in online coding challenges.
Worked on a new technology which enables cross-platform fragment and experience sharing, saving time and increasing code reusability across all Salesforce properties. Worked on core Salesforce technologies such as Lightning Web Components, Flexcards, Flexipages and Lightning App Builder.
Worked on detecting cyberattacks and classifying them using Machine learning and Deep learning.
Worked on implementing models for career path and job role recommendation systems.
Implemented a robust image segmentation model for anomaly detection.
Developed an objected detection model for advanced driver-assistance systems to perform real-time detection and recognition of objects.
Re-identification of a person across an array of cameras, which have a non overlapping field of view using state of the art machine learning models such as GAN and using GAIT-analysis (used for biometric recognition ).
View codeWorked at Robotics Research Group, IIT BHU trying to achieve swarm intelligence in robots using multi-agent reinforcement learning. I mainly implemented the clustering, representation of kilobots and Reinforcement learning algorithms for the swarm system.
View GymFourth year, Electronics and Communication engineering. CPI: 9.19
Built a pipeline to input two images obtained from stereo imaging to produce a detailed map of the image, which includes identifying the objects, their position relative to the observer and the distance from the observer. Obtained distance of objects from stereo images using triangulation and stereo depth mapping techniques.
View ProjectDetecting license plates using yolov5 object detection models. Using opencv to perform image segmentation to extract characters from license plate and training ResNets to classify them. Also implemented an end-to-end number plate recognition model using yolov5s and yolov5l.
View ProjectWorked on implementing RL algorithms such as Q-learning , Deep Q networks, Deep Deterministic Policy Gradients and Actor-Critic algorithms on basic games such as atari breakout games .
View ProjectOCR of hindi language characters using opencv to segment out characters from words and ResNet models to classify segmented characters. Classification accuracy was 99.72 on the validation dataset.
View ProjectBuilt a skin lesion classifier using Pytorch and densenet121 architecture on the HAM10000 dataset. The top accuracy achieved on the validation set was 0.906.
View ProjectTrained a model to identify the letters and numbers hidden captchas. The model was built on a faster_rcnn_inception_v2_coco object detection architecture and using the Tensorflow object detection API. It had an average validation accuracy of about 99.2% on all the characters.
View ProjectBuilt a chest radiology classifier using Pytorch and Resnet34. Top accuracy achieved on the validation set 0.95
View ProjectBuilt a model to classify handwritten bengali characters based on their consonant diacritic , grapheme root , vowel_diacritic using Convolutional neural networks followed by fully connected neural networks using tensorflow keras and open cv.
View ProjectBuilt a sentiment classifier on Smile Annotation dataset using pytorch and BERT pretrained model.
View ProjectNLP model built using BERT (Bidirectional Encoder Representations from Transformers) to classify tweets. Using a data set on tweets to classify tweets pertaining to any disaster . It got a validation F-score of 85 and training F-score of 89.
View ProjectTrained a model to classify chest radiology images of potential COVID-19 patients. The model consisted of Convolutional neural networks followed by fully connected neural networks using tensorflow keras and sklearn . Data augmentation was used to further improve the accuracy. It received a test accuracy of 96.67% and a training accuracy of 94.62%.
View Project