Inspiration
Shipping quietly moves almost everything we use nearly 90% of global trade travels by sea yet the way ships choose their routes hasn't really changed in decades. Most vessels still just take the shortest line between two ports, ignoring the weather, currents, and waves they're going to sail straight into. That struck us as a strange blind spot for an industry this massive. We started digging into it and found that this "blind spot" isn't small shipping accounts for close to 3% of global carbon emissions, and a huge chunk of that comes down to inefficient routing that burns more fuel than it needs to. Once we saw that gap between "what's possible with data we already have" and "what's actually being done," we knew we wanted to build something that closed it.
What it does
Our platform recommends fuel-efficient shipping routes instead of just the shortest ones. A user picks a departure port, a destination port, and some basic vessel details, and the platform generates two routes side by side on an interactive map: the traditional shortest-distance route, and an AI-optimized route that accounts for real weather, ocean currents, and wave conditions along the way. Underneath, a machine learning model predicts how much fuel the vessel will actually burn under those conditions, and a route-optimization engine uses those predictions to find the path that costs the least fuel not the least distance. The platform then converts that fuel savings into an estimated reduction in CO2 emissions, so the user sees one simple, honest number: how much fuel and how much carbon this route saves compared to the standard one.
How we built it
We split the problem into two halves and tackled them separately before wiring them together. First, we modeled the ocean as a weighted graph rather than open space each segment between waypoints carries a "cost," and instead of that cost being distance, it's predicted fuel consumption. Second, we trained a machine learning model on voyage data (weather conditions, vessel speed, cargo load, wave height, current direction) to predict that fuel cost for any given segment and vessel state. Once the model could estimate fuel burn reasonably well, we fed its predictions into a shortest-path-style search algorithm, so the "shortest path" it finds is actually the cheapest one in fuel terms. On top of that, we built a simple web interface where a user can pick ports and vessel details and see both routes plotted on a map, along with a summary panel showing the fuel and emissions comparison.
Challenges we ran into
Getting real, clean maritime weather and vessel fuel-consumption data on a tight timeline was harder than we expected a lot of the good datasets are scattered, inconsistent, or require access we couldn't get quickly, so we had to be resourceful about combining what was available with realistic simulated data without making the results feel fake. Balancing the machine learning side with the optimization side was also tricky: it's easy to over-invest in one and leave the other feeling like an afterthought, so we had to keep going back and forth to make sure the fuel-prediction model and the route engine were actually talking to each other in a meaningful way, not just bolted together. And translating something technical a graph search over predicted fuel costs into a demo that's genuinely easy to look at and understand in under a minute took more iteration than we expected.
Accomplishments that we're proud of
We're proud that we got a full pipeline working end to end from user input, through weather and fuel prediction, through route optimization, to a visual comparison on a map rather than stopping at a partial proof of concept. We're also proud of how clearly the final output communicates its value: seeing two routes side by side with a straightforward "this saves X% fuel and Y tons of CO2" is the kind of result that doesn't need a long explanation to land, and that clarity took real work to get right. Beyond the build itself, we're proud we picked a problem that's not just a cool technical exercise it's tied to a real cost and a real environmental impact that the shipping industry actually cares about.
What we learned
We learned a lot about how much nuance goes into fuel consumption modeling it's not just speed and distance, but a genuine interplay of currents, wave resistance, wind, and cargo weight, and even a simplified model taught us to respect how complex real maritime operations are. We also learned how important it is to design for interpretability, not just accuracy a black-box prediction is a lot less useful to a ship operator than one that can be tied back to "this route has rougher seas, so it costs more fuel." On the team side, we learned how valuable it is to prototype the end-to-end flow early, even roughly, instead of perfecting one component in isolation having something working start to finish early on made every later improvement easier to test and justify.
What's next for AI Shipping Route Fuel Optimizer & Emission Reducer
The next step is replacing our simulated data with live AIS vessel tracking and real-time weather feeds, so the recommendations reflect actual conditions rather than historical patterns. We'd also like to evolve the route engine from a static optimization into something closer to a reinforcement-learning agent that can adjust a route mid-voyage as conditions change, rather than only planning once at departure. Beyond a single ship and a single voyage, we want to scale this toward fleet-level optimization helping shipping companies plan and schedule multiple vessels at once and eventually add features like port congestion awareness and automatically generated emissions reports, so the platform isn't just a routing tool but a genuine operational and compliance asset for the industry.
Log in or sign up for Devpost to join the conversation.