Working on large scale Personalized Recommendation systems for Samsung's Ad targetting team.

Smart Troubleshooting for service incidents:  Developed an NLP and Machine Learning based product for diagnosing issues of advanced medical devices (like CT,MRI) and recommending troubleshooting steps to service engineers, which led to productivity savings of around $50 million per year. The product involved the following major sub-projects:  *Formulated and developed a novel (patented) semi-supervised data labelling process using an ensemble of diverse classifiers followed by feedback based (active learning) process to build reliable labelled training data *Developed two tier hierarchical prediction model (multi class classification) which pin points the machine issue *This involved incorporating large volume (>500 GB input datasets) data-sources(with very low SNR ratio) and processing semi and unstructured datasets (user written text data, machine log data, sensor based log data) *This process also involved modelling for high cardinality and high class imbalance using effective sampling and hierarchical classification with uncertainty thresholds *Created custom model explainability models (based on LIME) for machine codes to gain the trust of the users towards our predictions and hence boosting tool adoption and end user confidence in predictions *Model performance monitoring framework with embedded feedback mechanism in the tool for active evaluation of model in the field *Scaled up the tool from a single region to all regions while handling linguistic differences on text model performance *Co-designed Model Ops with data engineering team to facilitate easy model deployment and management in production in AWS platform *Recruited and mentored junior data scientists to create deliverables for the team *Created several POCs for ML evangelism across different orgs in GE (Some prominent examples include Graph based supply chain optimiser, Product Reliability feedback text clustering)

Data Scientist at General Electric Healthcare.Working on problems with an NLP focus.

At Mu Sigma, I worked as a Decision scientist who works at the intersection of Math, Business, and Technology to solve complex business problems using Data. I have worked on a broad field of problems ranging from Extracting information from Text (NLP), Images(CNN, SSD) to predicting purchase patterns of customers ( Clustering, Regression) using a variety of tools like Python, R, Hadoop, Hive, Oozie.Key Projects: Customer Behavioural Segmentation: Enabled a major US retailer to make better marketing and merchandising decisions by segmenting their customer base of ~90M households based on their purchase pattern Omni-channel Propensity analysis: Increased Email CTR (Click Through Rate) from 1.6% to 3.1 % for a major US retailer among traditional Brick & Mortar shoppers by prioritising potential Omni-channel households through a look-alike model Customer Lifetime Value model: For a US retail major, created a CLTV framework to predict the total spend of a customer for the next 3 years. The framework was built across different tenure-based shopper groups to enable better marketing, using a suite of regression models along with migration probabilities.  Fruit Defect detection: For I & D wing of a leading US retailer, built fruit defect detector and classifier using Deep Learning techniques(r-CNN, SSD) which predicts defects in Strawberries with an accuracy of 93.5%.