Session-Aware CV Recommender System - Built a keras model to predict vector representation of a CV in latent space by using textual & numerical information. Models the preferences of a recruiter at session level by observing sequential behavior using LSTM.

Aspect based sentiment analysis - Integrated with main platform, leveraged by multiple teams for sentiment analysis and used by over 1200 Sprinklr clients. Built sentiment analysis model at target word level for all Sprinklr clients supporting European languages; achieved an accuracy of 80% (jump of 25% over existing model) using Recurrent Neural Networks.Twitter profiles demographics prediction - Integrated with main platform, leveraged by listening module for audience insights and used by over 1200 Sprinklr clients. Developed a plugin framework for predicting insights on all twitter profiles of over 800 million active users. Applied transfer learning on Google’s inception model for predicting inappropriate content on images, used for categorising twitter profile spam and predicting age category and gender of twitter profiles; achieving 82% accuracy. Trained a Conditional Random Field model to predict marital status, parental status of twitter profiles.Influencer Score Prediction - Built a ridge regression model for predicting social profile’s influence using social profile metrics. Applied feature selection on profile metrics to extract most correlated attributes, finally achieved r2 score of 0.64.Video Similarity - Predicting similarity between video files based on the content, using nearest neighbours between video frame vectors. Perpetual hash vectors used as video frame vectors, nmslib is used for finding vector similarity.

Research Internship at Sprinklr R&D, where my work included -1. Predicted industry specific attributes relevant for increasing popularity of brand posts on social media pages2. Built a hyper parameter tuned SVM model for predicting brand post popularity; achieved accuracy of 86%