Media Buy Clearance Prediction – Project Lead
o Drove cross functional collaboration among engineers, UI specialists, and business stakeholders to improve clearance prediction model and enhance the analytical insights on client’s buy management platform for buyers;
o Quickly iterated underlying prediction model and its deployment through python to Watson Machine Learning;
o Demoed prototype of visuals on how buyers utilize model predictions to decide best unit price of bids and best daypart to place ads by Jupyter interact widgets;
o Mentored junior data scientist on end-to-end model development pipeline;
o Received satisfactory feedback from client. They'll continue and double software subscription soon.
Continuous Monitoring of Productionized models
o Designed architecture to embed fairness monitor, explanation provider and drift monitor of Watson OpenScale to evaluate live model performance from different perspectives; Implemented through python to Azure ML;
o Successfully set up new monitors within client’s existing production pipeline for smooth operational change, resulting in 27 times software deal growth.
Data Quality Auto-Detection
o Identify key use case as contextual anomaly detection out of ambiguous descriptions from client;
o Co-developed a method to detect contextual anomaly without domain knowledge infusion using PySpark on Databricks;
o New method outperforms common algorithms on KDD Cup 99 data. Pending patent approval.
Helped IBM develop strategic relationships with various clients
o Built end-to-end data science workflows and presented them in meetings with clients to show value of IBM data science products
o Joint data science discussions across different industries including retail, finance and media buy as data science expert. Provided vanilla solution prototype to our client during the meeting
