Unsupervised Machine Learning Coursework

Experience

For a healthcare client, devised Covid-19 data lakes using audit tables for data validation with custom email alerts. Generated optimized and scalable data extracts on patient data of over hundred million records using Scala with Spark, Hadoop and Hive.
Constructed cost reduction, vendor consolidation tool using Django. Developed and deployed the tool with backend xgboost models onto GCP, allowing users to train-test with a goal to classify healthcare purchase orders, choose the optimal pricing options and reduce manual efforts saving ~18% costs annually.
For a pharmaceutical client, generated scalable feature engineering pipelines to create a 360 degree patient view using Airflow and PySpark. Conducted hypothesis driven testing at a patient/drug level over clusters to help client better target the right population. Built Uplift models to test various health plans with a goal to identify the success rates across treatment groups, improve adherence and scripts.
Created Data cleaning framework for in-house ML package with end to end testing. Built pipelines to auto deploy transformation models onto MLFlow for reusability, optimizing the build time.
• Conducted internal AI hackathons for practitioners across the firm. Contributed to University related recruiting info session talks, studio jams and career fair events.

Automated an audio-based customer feedback tool on predicting emotion from speech features and its text translation using LSTM and CNN algorithms with Keras and TensorFlow.
Researched on analysis of fake reviews and their impact using Near-duplicate detection, LIWC based classification with Map Reduce techniques on Amazon reviews Dataset with AWS S3 and EMR.

Automated the Data platform to pull data and map it from external vendor API’s. Integrated product booking, payments into the Placepass engine API platform with micro-services using .Net and Java.
Led and owned Machine Learning workstream to classify multi-labeled products into the site catalog. Generated tags words for content with LDA/NMF topic modeling using NLTK, SK-learn and Python.

Developed a Graphical Web Application using MEAN stack which visualizes data logs from flights by multiple profiles, assisting maintenance engineers and replacing manual efforts drastically.
Rapidly prototyped a POC web app to assist pilots with various navigational protocols while performing search and rescue operations during natural disasters using LAMP stack and Google Maps API.

Researched on Emotion Prediction in Email Conversations using Natural Language Processing, which required me to implement a spam classifier and then predict emotion from ham emails, using Java with Apache Lucene and Stanford NLP libraries.