Microsoft CopilotMicrosoft news and feeds recommendation

Yahoo Research | Yahoo Knowledge GraphInformation Extraction• Continuous Learning, Multi-label classification, Convolutional Neural Networks, Fine-grained entity typing, Weak Supervision, Topical extractive summarizationWorking on Information Extraction from Sports news articles to predict prevalent rumors.• Developed a method for Topical Extractive summarization of rumors in NBA, NFL sports articles• Implemented Weak Supervision techniques to increase labeled data• Implemented Convolutional Neural Network(CNN) to perform multilabel classification of sentences into predefined categories in Python• Implemented the concept of continuous learning to avoid retraining the model from scratch on the arrival of a new topic data• Developed a framework for finding out related entities and fine-grained entity typing.Video Intelligence:• Single Shot Detector Model, Convolutional Neural NetworksWorking on generating a fine-grained action recognition dataset for NBA, NFL, NHL, and Soccer sports using their broadcast videos. • Developed an end-to-end system to map the above said sports game scenes to text commentaries from Yahoo! Sports API by accurately recognizing the text from the clock.• Implemented a Single Shot Detector(SSD) model to detect the bounding box around the clock in the sports broadcast videos. Passed the extracted bounding box for text detection and recognition• Generated ground truth labels for different moves in the above-mentioned games by matching the time recognized in the clock with Yahoo Sports API, for enhancing user experience.• Benchmarking the dataset using l3d and r3d action recognition frameworks.

• Graduate Teaching Assistant for the graduate level course CSE 512 Machine Learning taken by Professor Minh Hoai Nguyen • Conducting office hours to address the queries of students and creating grading rubrics• Proctoring exams, manual grading, leading meetings and coordinating assignment development along with the Professor and 2 tutors

• Predicted the time required to resolve an issue, by processing natural language in emails using LSTM.

• Developed a framework for categorizing customer queries from a chatbot using Hierarchical Temporal Memory (HTM) Theory in Python. • Used Latent Semantic Analysis for generating word embeddings.• Applied HTM theory for learning and noise reduction • Implemented a single layer feed forward neural network for classification.• Improved the classification accuracy of the company data by 28% and achieved an accuracy of 83.19% on the 20 newsgroup dataset.