Advising
Since 2024 I have advised MS in Business Analytics capstone teams working with companies and nonprofits, and students doing independent studies. 13 projects so far, described here without the students' names and without the partners' data. Starting in Spring 2027 I also advise capstone teams in the MBA Business Analytics track.
Partners include Westinghouse, RBC Wealth Management, Swank Construction, SaratogaRIM, Confirmed, Marinus Analytics, 412 Food Rescue, Tindoori Labs, and a large consulting firm.
Advised
Microreactors cost more than grid power but keep running when the grid does not. The team forecast diesel, electricity, and natural gas prices by state and sector through 2038, estimated what outages cost different customers, and built pricing dashboards for government and commercial buyers.
Thorium fuel has real advantages, but no U.S. commercial program exists. The team built a decision model that combines the odds of development success, operating costs, and net present value, then ran it across dozens of scenarios using only public data to find the conditions under which a project is worth pursuing.
Universities are adopting AI analytics tools without evidence that they help. This IRB-approved pilot experiment compares a conventional dashboard with a natural-language AI interface on the same admissions data, measuring accuracy, confidence, trust, and cognitive load.
Default stocking levels in hospital dispensing cabinets cause both stockouts and waste. The study built a pipeline that turns a year of cabinet transactions into recommended levels, compared optimization against simple rules in a replay simulation, and found that how restocking happens matters as much as the levels themselves.
Earnings announcements push option volatility up and then crush it. The study back-tested volatility-only option strategies on 30 heavily traded stocks over five years and found that selling volatility worked where buying it did not, unless entries were filtered carefully.
Advised
Swank priced state highway bids mostly on experience and was losing close ones. The team scraped five years of public Pennsylvania DOT bid results, over 750,000 line items, and built a dashboard that benchmarks unit prices against winners and competitors by project type and region.
The team segmented U.S. households using Census wealth data, analyzed more than 12,000 posts about wealth transfer with sentiment and topic models, and recommended a two-track engagement strategy: one for the parents passing wealth on and one for the millennials receiving it.
Instead of learning after the fact that a managed account had drifted from its stated strategy, the firm wanted a warning in advance. The team built factor exposures for every portfolio, defined drift, and compared tree ensembles with neural networks for predicting it over one, three, and twelve months.
A scheduling software company wanted to know whether a 14-day or a 30-day free trial works better. The team ran a randomized survey experiment with 300 sales professionals, found that trial length did not change willingness to start but did change whether people felt they had time to decide, and recommended a two-track trial.
Click data shows what users do, not why. The team built a system that reads a live website and has language models review it as about 25 detailed personas, including users who rely on accessibility features, then checked how consistently different models rated the same site.
Advised
Marinus Analytics helps law enforcement find human trafficking and already flagged ads with an image model. The team added text models to confirm high-risk ads and cut false positives, comparing classic features, embeddings, and a fine-tuned transformer.
A small share of scheduled food rescues were never picked up, wasting food and frustrating donors. The team modeled which rescues were at risk and why, and pointed to practical changes in how recurring and one-time donations are posted.
An early-stage app pairing Asian-language learning with relationship building needed to know who would use it. The team ran a bilingual survey with nearly 400 respondents and recommended a target segment, a pricing model, and the privacy and quality controls respondents cared about most.
Before Tepper
Independent studies I advised while teaching organizational behavior, mostly at Carnegie Mellon's campus in Qatar.
