Titas Das

Customer Loyalty Engine: Measured and predicted customer loyalty trends using a combination of rfm metrics, causal graphical models and a plot based algebraic function. Detected products and specific product features that could induce loyalty in customers. Insights from Images PoC: Generated captions detailing the interaction between a product and its environment using object detection, person detection, brand detection, scene understanding and image captioning. Detected attributes in clothing. Eg - type of material and categories of clothing using deep learning. Product Recommendation Engine: Used attributes, categories from the above module and additional features to develop a product recommender that could bundle and promote products (used for a client that was in the clothing and fashion industry). Optimized the recommendation engine for novelty and diversity.

Started the labs division which was responsible for suggesting ways in which existing features could be improved and the product portfolio could be diversified. Recommendation System for Financial News: Developed a hybrid (content and collaborative filtering) recommendation system using customized features. Prototyped an application in flutter to serve this recommendation system to end users with the goal of improving accuracy and novelty of the recommended content. Stocker PoC: Designed and developed a system capable of detecting events in news articles and linking related events, thus making it easier for the user to follow a news story. Text Summarization: Implemented (extractive and abstractive text summarization approaches) from papers along with developing a pipeline with news collection and article filtration. Leveraged summarization to give users a quick overview of news articles. News Section: Reduced clutter on the user's news feed by getting rid of duplicate news articles using automated k-means clustering and a modified latent semantic analysis algorithm. Price Prediction for automated algorithmic trading PoC: Experimented with random forest and LSTM networks to predict share prices. Customer segmentation and analytics: Validated target groups for new features using hierarchical clustering and DBSCAN.

Care Drivers - provided services like ridesharing/pickup-drop and daycare for kids. Designed the architecture and pipeline for the application as part of a two person team. Devised an optimized driver matching algorithm using a combination of time/distance and area coverage functions. Developed the android version with features such as ridesharing, livestreaming, communication services (twilio API) and payment services (stripe API). Introduced a safety feature using facial recognition to alert the parents when the child is stressed or scared during the ride.

Research Objective: Predict an user's political standpoint and the extent to which they are able to influence their network. Used multi-kernel support vector machines to classify the behavior of users and predict their chances of sharing content on social media indicating their inclination towards a certain opinion/viewpoint. Analyzed layers of an users social network and looked into likes, comments, choice of words and sentiment to understand an user's political inclination and the likelihood with which they were to support a particular opinion or viewpoint at present or in the near future (within 6 months).

Competitive Cue Sports - Android Engineer: Developed an android application for local players to set up matches and engage in competitive cue sports. Client who was a former professional pool player was able to scout for local talent and train them through this application. Shot Tracer - Computer Vision Engineer: Improved an already existing ball tracking algorithm to track the speed and force with which a baseball is hit for an application that was designed to help train for baseball.

Feulin - An android application that made everyday cooking enjoyable. Developed a recipe recommendation system using features based on taste, meal type, expiry date. Developed a novel taste based metric to refine the recommendation algorithm. Integrated OCR using LSTM networks to read from receipts.

Relevant Courses: Advanced Machine Learning (Measure Theory, Optimization Theory), Advanced Machine Intelligence (Natural Language Processing), Image Processing and Computer Vision, Pattern Recognition (Statistical Learning Theory), Big Data in Ecosystems (Deep Learning, Distributed systems), Math for Intelligent Systems (Real Analysis, Vector spaces, Linear Algebra, Probability Theory, Information theory), Computational Neuroscience (Sensory encoding and decoding).