On Monday, November 10, 2025, the Center for Smart Streetscapes (CS3) welcomed one of its earliest contributors, Dr. Patrick Alrassy, for an in-depth discussion on his interdisciplinary career trajectory, the foundations of his research, and what it is like to work on Artificial Intelligence at Meta. Although he completed his PhD two years before CS3 was formally established, his work played a pivotal role in securing the center’s initial funding. By demonstrating that large-scale urban mobility analytics could produce meaningful real-world outcomes, Dr. Alrassy helped validate CS3’s mission during their first large data project.
Building a Career on Interdisciplinary Curiosity:
Dr. Alrassy joined Columbia University’s Department of Civil Engineering in 2017 as a recent PhD. Graduate. Working alongside Professor Andrew Smyth, PhD. – CS3’s Director – and Professor Jinwoo Jang, PhD., he joined a landmark collaboration with the New York City Department of Transportation (NYC DOT). The project focused on citywide telematics data collected from over 27,000 municipal vehicles, with the broader goal of modeling driver behavior and identifying new pathways for improving roadway safety.
An early internship in structural design introduced him to applied engineering in Manhattan, but he quickly realized he wanted to pivot his career and looked towards Columbia University. After contacting Professor Andrew Smyth, he joined the research team working on the New York City Department of Transportation’s large data project. He soon became an integral part of the team, helping solve some of the largest issues facing their research. This experience allowed him to realize his passion for artificial intelligence, and set the course for the trajectory of his career.
Following his time working with what later became the Center for Smart Streetscapes, Dr. Alrassy moved through a series of high-impact roles – first at the startup Optimus Ride, then to Meta’s Oculus division, where he contributed to safety systems and high-fidelity image-mapping technology. Two years ago, he transitioned to Meta’s Superintelligence Lab, where he worked on the evolution of large scale AI models, including LLAMA Models.
Contributions to Research at the Center for Smart Streetscapes:
During his time at Columbia, Dr. Alrassy contributed to several core research outcomes that remain active in New York City’s operational infrastructure today. Among his most influential contributions were:
- Citywide traffic speed distributions (Speed Enforcement Initiatives/Deployments)
- Developed a web application that is used internally within NYC DOT that returns the road safety metrics of New York City Roads
- Created a map matching software that maps ambulance trajectories with GPS points to a digital road network
- Generated synthesized predictive travel times that have been a pivotal input in ambulance travel modeling that was also used as part of the Covid-19 pandemic hospital load balancing algorithm.
- Optimized ambulance routing in NYC, lowering the traveling time it takes from incident location to hospital by an entire minute.
These accomplishments emerged from an exceptionally challenging dataset. The research team worked with noisy GPS readings captured at 30-second intervals, incomplete map references, and sparse sensor-derived speed and braking metrics. Before any behavioral modeling could begin, the team had to reconstruct the underlying trajectories of tens of thousands of vehicles across the city – which was no small undertaking.
How Dr. Alrassy Helped Solve the Map-Matching Problem:
A central scientific challenge in the project was accurate map-matching—the process of determining which road segment corresponds to each GPS reading. Because the DOT maintained its own custom roadway map, traditional commercial map-matching algorithms were insufficient. Dr. Alrassy approached the problem through a probabilistic framework:
- Emission Probability: Estimating the likelihood that a noisy GPS point originated from a given roadway segment using spatial distance, heading, and related metrics.
- Transition Probability: Modeling how likely it was that a vehicle moved from one candidate segment to another based on distance constraints, travel time, and on-board diagnostics (OBD) speed data.
This probabilistic structure was then reframed as a Dijkstra least-cost path algorithm, allowing the team to infer the most probable route given the noise characteristics of the data. After three years of refinement, the system proved robust enough to support the DOT and FDNY in operational decision-making, ultimately proving that there was a need for the research that CS3 would undertake.
Early Career Lessons from Dr. Alrassy’s Experience:
Reflecting on his time as a student, Dr. Alrassy emphasized that his career was shaped less by specialization and more by deliberate breadth. By exploring structural engineering, data science, AI, and urban systems, he gained the flexibility to transition across different fields and industries.
He noted several career insights for emerging engineers and scientists:
- Interdisciplinary training will help you broaden your skills and explore your passions.
- Start applying for jobs at least 5 months before you graduate, and apply to as many jobs as you can. He submitted 60 applications and interviewed at many different companies with as many as 7 rounds of interviews.
- Prepare for coding interviews by studying material, and practicing coding problems on LeetCode.·
- AI will not replace jobs; it will just change job titles. He suggested that titles may shift (e.g. from Mechanical Engineer to Computational Design Scientist), but the underlying need for human-directed problem solving will remain.
About Dr. Patrick Alrassy’s Work at Meta:
During his time at Oculus, Dr. Alrassy worked on Virtual Positioning Systems, which are technologies that ensure accurate spatial alignment in alternate-reality environments. These systems enable reliable object placement, environment boundary recall, and consistent user localization—which are critical components in the safety of Virtual Reality.
Two years ago, he transitioned to Meta’s Superintelligence Lab, an environment he described as “the most intense yet very fun.” During his time there, he contributed to the design and evaluation of advanced AI systems, including internal benchmarks for pretrained language models.
About training AI models, he advised listeners, “Develop and maintain internal pretraining evaluations designed to measure model capabilities in a controlled, uncontaminated environment—entirely absent from the public internet, to avoid training leakage.”
