Bot should be able to understand instructions given to them by the third person (dialogue designer) and generate utterances of desired task types (ex. say, ask, set, etc), for the exchange of information with the user in conversation.We created the dataset for generating dialogue responses from third person instructions, and fine tune T5 and BART models.

Worked on InSync, SCOOT, SURTRAC like traffic control system from scratch which follows NTCIP standards, currently the system is able to reduce waiting time up to 60 seconds in a cycle at a junction reducing the travel time, air and noise pollution for smart cities. The whole system will be able to provide a green corridor during peak hours using state-of-the-art AI algorithms.

To address the difficulty of obtaining good features, as well as improve system performance, we learned representations from signature images, in a writer independent format, using Convolutional Neural Networks, transfer learning with pre-trained models, DCGAN, which is followed by writer dependent classification which leads to FRR 3.95 and FAR 11.34.

Created an application for tablet-based digital ordering menu using Django Oscar-e-commerce framework for restaurants with the customer, chef, and manager interface, the aim is to predict the sales, inventory needs, etc. based on collected data with machine learning models.

Identified speaker in video using YOLO object detectionalgorithm and convolution neural network and transfer learning with VGG16, InceptionNet, ResNet, achieved testaccuracy of 75.16%.

Determined the word boundaries in new wav files given the list of wordsthat are spoken and their index with a sequential model (LSTM), achieved test accuracy of 60%.