• Part of the team responsible for the design and development of risk suppression models contributing to the direct mail marketing and credit card acquisition strategies in the shifting environment; resulted in reducing exposure to delinquencies (~70%) and increasing profitability in 22Q4. Implemented production-level code in Spark/PySpark to score over 220 MM records per cycle. • Developed extensive monitoring tools to ensure the stability and validity of developed models by tracking short-term proxy risk metrics over different segments and schedule model refresh cycles. •. Developed a novel framework to detect reliable attributes for model building after channel migration and minimize biasness by selection of optimal imputation strategy.

• Designing Hometap’s next generation of underwriting risk model (PD): Deployed efficient ETL pipelines, stratified resampling, and variable selection using Pyspark. Developed machine learning stack to improve Precision-Recall AUC by 13% which resulted in 8% lower disqualification rate at inquiry stage. • Created a probabilistic framework for model optimization attuned to the fund structures, conversion likelihood, and simulated investment performance. Resulted in potential 41% net revenue uplift per lead while adhering to risk guidelines. • Developed multiple propensity and survival models (time-to-sale, time-to-refinance, etc.) to predict contract’s time to settlement. Derived a parametric model to explain competing settlement mode under different hypothetical scenarios. • Devised a novel similarity-based algorithm to map risk profiles to the external data (Experian) for back scoring. Implemented a time-series clustering model to identify major market paths in CoreLogic’s home price index data. • Collaborated with the finance team to ensure model soundness and develop execution plan. Presented to the full chief executive team. Delivered technical reports to answer the company’s investor’s questions (Bain Capital, General Catalyst, ICONIQ).

• Developed an end-to-end machine learning stack for real-time borrower risk rating: Analyzed 30+ GB of data from origination, sequence of payments, and call center conversations content and other alternative datasets. The model reached 0.76 ROC-AUC, with a lift of 3.71 among the top 20 percentile of scorecards. • Conducted quantitative market research on internal & third-party data in response to proposed sales & pricing strategies • Partnered with IT team to maintain data pipelines (Oracle, AWS-EC2) & delivered models in live production via RESTful API. • Created dashboards to visualize the personalized breakdown of scorecards (Shapley value) for the operation teams.

Statistical Learning Approach to Determine Optimal Sizing & Investment Timing of Commercial-Scale Distributed Energy Resources

The first sport-based social network currently ranked under 100 by Alexa in Iran