Backend engineer for the Microsoft Teams Extensions App Store, a platform where developers can create, upload, and manage apps for personal, enterprise, or global use, and end users can install these apps to enhance Teams, Outlook, and Office functionality. Built using C# and Azure, the platform supports over 50M+ DAU, 3,000+ 3P extensions, and numerous enterprise apps.Previously, Owned the Microsoft Teams Call Queue Service, which serves over 8M calls per month, and is used by enterprise customers to set up call routing mechanisms for their customer call centers, built primarily with C# and Azure

Conducted Research on Reinforcement learning (RL) at the Media and Data Science Research Laboratory, Adobe1. Worked on extending state-of-the-art RL policies using Adversarial training2. Developed a Regularization technique for Domain Randomization to help learn more robust and adaptive locomotion policies

Worked on a recommendation task for the product Microsoft Teams.1. Employed machine learning models to compute user-channel (i.e. item) affinity scores and to perform the recommendation task(Using python, Scope, TLC, cosmos)2. Experimented with different label aggregation methods, features, time decay, models, etc. and performed analysis on the training data and the results of modelling3. Generated user-level and channel-level statistics using the generated affinity scores4. Integrated the models, computations and scores into the data pipeline of the product using Scope, TLC into Cosmos

Implementing state of the art deep learning techniques for various text classification tasks using tensorflow and python1. Trained word embeddings using Tomas Mikolov et al's word2vec continuous skip-gram model2. Classified text documents into fixed categories, sentiments using various techniques such as simple feed-forward architectures, CNN, RNNs3. Classified text documents into dynamic categories( provided at runtime) implementing several techniques such as Doc2Vec, Averaging, cosine similarity measures etc4. Implemented Long Short-Term Memory(LSTM), an RNN architecture to classify movie reviews to movie genres5. Studied and analysed the effects of retrained word vectors on classification tasks