2024 MIT R&D Conference: Startup Exchange Lightning Talks - Mobi Systems

PENG YU: Hi, good morning, everyone. I'm Peng Yu, from Mobi Systems. Really glad to have the opportunity to meet all of you today.

So Mobi is also a spin-off from MIT, actually from the computer science and AI lab right across the street. So today, many of the teams still actually have the route back from the AI lab.

At Mobi, we develop a large-scale planning and scheduling system that can really solve for large-scale planning problems with uncertainty, and more importantly, in collaboration with humans. And then secondly, it's not just generating a static plan, but able to quickly adapt to any changes, disruptions in a highly dynamic environment. So you can probably already guess, actually, our first vertical to start delivering our product is in travel and transportation.

Our launch customer is TUI, from Hanover, Germany. They are the largest tourism group in the whole world. We are basically, the core among their platform to operate transfer services across their worldwide business.

This kind of illustration gives you an idea of how the system works, where effectively at the center because among all the different parties, from the travelers to their operation team to the drivers operating the vehicles, that coordinates the dispatch, the decision-making, the resource allocation among all of them. Hundreds of millions of decisions flow through our platform every year, and you can imagine from this diagram, each party is receiving different information from the system, as well as providing additional constraints input back to us to request updates or new plans to be generated.

So the operational efficiency, as you may guess, is actually quite straightforward. It actually comes in three folds.

First, we will actually be able to greatly lower the planning time, from hours down to minutes and then sometimes in seconds, allowing a way more agile operation. Secondly, because we are a central platform connecting all of the different parties and equipment operating in the network, we're able to enhance the decision-making, looking across all the constraints not requiring humans to connecting different pieces together.

And finally, we are also achieving greater flexibility, supporting more than 1,000 different types of vehicles, and even more types of business constraints, to allow the system to perform differently for different countries and different destinations.

The past 12 months is the first time we achieved a global rollout, and we were able to support more than 25 million passengers throughout the 12 months. As you can imagine, behind the scenes, actually hundreds of millions of decisions being made automatically by our system. And then even during a peak day, more than 200,000 passengers are supported by our platform for their journeys through the TUI network.

This platform has also been proven to be robust across 440 different countries, more than 100 destinations, and we have successfully achieved more than 99% reduction on planning time, and a cost reduction from the very first day we turned on our platform.

Moving to the next stage, we also recognize the complexity of deploying such large-scale, complex planning systems with thousands of parameters and constraint types. So really, in the past two years, large language model development has really given us an edge to actually tackle that, by allowing an automated system to work together with the operators of our system, to actually give them a natural language interface, and then behind the scenes, automatically map what they're looking for into Structured Query languages, describing the constraints, the objective functions, such that people without much training, they can still be working with the system on day one.

Let me give you a very simple example, like a search query. I want to look for a Ritz-Carlton by the ocean that's also warm in January in a city with nonstop flights from Boston. You can imagine, you have to tap into many different systems, many different search, routing, transportation, planning functions, and then distributed data set in order to come up with the accurate answer. But now, with our natural language interface and then really the already powerful platform, everything can be handled in real time and in a fully automated way.

So the reason we're here is we believe the planning, scheduling, and resource allocation platform we have developed would have huge impact beyond travel, and all the work we have done before may have a similar impact in maintenance repair, disaster response, transportation, or manufacturing and supply chain whenever we have to make decisions in collaboration with humans for space, for time, and resource allocation.

We're looking forward to share more and learn more from you during the lunch break after this session. Thank you.

[APPLAUSE]

SPEAKER: Thank you so much, Peng.