Some problems need more than an API call. Pricing engines that underwrite billions. Computer vision that earns FDA approval. Predictive systems that win championships. Dr. Erik Allen and team have been solving these for over 20 years — and shipping them into production.
A selection of projects where we’ve partnered with companies to solve their hardest problems.
Designed decision-making software used by pit crews during live races. Contributed to three NASCAR national championships. Technology acquired by General Motors (NYSE: GM) in 2020.
Worked with Allianz Group Technology to build Project Nemo, their first agentic AI system for automating insurance claims. Reduced claims processing time by 80%, settling simple claims in minutes instead of days. Built and deployed in under 100 days.
Built a computer vision model for lung cancer detection that surpassed FDA medical device standards, leading to 510(k) FDA approval for Imidex's VisiRad XR product.
Read case study
A team of MIT PhDs built statistically rigorous predictive models for baseball betting, years before analytics-driven wagering existed. The models returned 33,000% in a single season on a small bankroll. The team later collaborated with Billy Beane, the Oakland A's GM behind the Moneyball story.
Architected the core loss ratio underwriting model for Workers' Comp, one of the hardest lines in insurance to price. The model now underwrites over $500M in annual premium and is the foundation of a billion-dollar business.
Built convolutional neural networks that detect and count waste containers on truck routes in real time, turning a previously opaque fleet operation into a computable problem. The models enabled remote monitoring of truck assets at national scale and became the foundation of Rubicon's Fleet Technology Business Unit. Sold for ~$100M in 2024.
Built a nationwide pricing engine that cut average quote time from over a day to under a minute, unlocking the growth trajectory that led to Rubicon's IPO.
Built the core data warehouse that unified OpenWorks's disparate data sources, then designed automated pricing models on top of it that brought pricing decisions down to sub-second response times across multiple systems.
Built convolutional neural networks for automatic vehicle recognition at parking facility entry and exit points. Replaced the need for attendants or mechanical gate systems with a single camera feed handling identification and access control. Enabled FlashParking to scale to over 16,000 locations across North America.
Combined sales data with weather, events, holidays, and TV schedules to improve demand prediction accuracy by 40%, directly reducing labor costs and waste across locations.
Built a system that scrapes 100+ government websites and delivers weekly AI-generated intelligence summaries, replacing days of manual research with a single automated pipeline.
Read case studyReady to start?
Tell us about your data science challenge and we’ll set up a time to talk.
