Worked with Canvas learning management system(LMS) research team to built various ML solutions for existing as well as new features.- Designed and implemented a personalized adaptive notification ML service called Nudge within Canvas that helped students efficiently manage their time and coursework by nudging them before the assignment was due.The idea of the project involved model free reinforcement learning(Cross Entropy Method) which was implemented using Multi Layer Perceptron in Pyspark.- Fully owned and implemented a sentiment analysis feature within Canvas insights(Learning analytics platform to promote student success aided by machine learning) which surfaces positive, negative and mixed sentiment on Canvas Discussions and Conversations data. The project also involved conducting user interviews by piloting with universities. Initial model was training on the Stanford MOOC dataset using Spark NLP bert_base_uncased pre-trained (1024 features) network followed by a MLP classifier in Pyspark. - Developed a baseline model for automating auditing a course for admins within Course Readiness(Tool within Canvas Insights for institutions to implement best practices across all courses) which helped in further optimization by saving admins approximately 6-10 hours per course.- Worked on improving features for Canvas Insights by analyzing Canvas students data to see if course design practices by teachers has any impact on student success. Performed t-test on bunch of features to judge the level of significance. - Lead and implemented the automation of a particular Support ticket Salesforce feature when analyzing the root cause of high number of Canvas support tickets observed during COVID-19.

- Worked on building an Automatic grading software which does partial marking by measuring the similarity between MATLAB source codes - Subsequence Matching algorithm for measuring similarity showed 99% accuracy on little tweaking of source code - n-gram feature engineering followed by applying latent semantic indexing/topic modelling lead to an 90% accuracy on little tweaking of source code- A deep learning algorithm by training the GRU using neural network matlab source code vocabulary to create a Code2vec model in the latent space and hence measure cosine similarity ,the accuracy observed was 87%- Automatic Grading System is being used for assigning Partial credits scores- Currently researching Abstract syntax based parsing technique for similarity measurehttps://github.com/richars7/Automatic_Grading_SystemMentor: Dr. Richard Golden

- Worked on exploring IBM-No Attrition and fMRI Big dataset by applying various multivariate techniques- Performed data compiling, cleaning and standardisation of IBM-No Attrition dataset and further applying inferential statistical techniques like PCA, MCA, PLS, BADA, DiCA, Di-STATis, MFA etc. to derive insights about the data- Presented the research analytics and published Self-Book on Multivariate Analysis using R- https://bookdown.org/richa/Cookbook/dataset-ibm-hr-employee-noattrition.htmlMentor: Dr. Herve Abdi