KITT AI Freight Optimizer | Production-ready AI Agents Hackathon @ AI By the Bay

KITT AI Freight Optimizer | Production-ready AI Agents Hackathon @ AI By the Bay

Recording: KITT AI Freight Optimizer | Production-ready AI Agents Hackathon @ AI By the Bay

All right. Pretty much projection, screen projection in Linux is kind of complicated. So, uh, initially I'll walk you through our idea and I'll explain it and then we'll figure out uh the other part. So, my name is Yab and this is my friend and teammate. My name is Sam >> and uh together we built a tool called uh kit. So uh so basically kit is a solution for a uh an industry that is estimated to be an $800 billion a year but the problem itself is a $12 billion a year cost. So what is that problem? So I am a firm believer that every single person in this room have ordered something online and I hope I'm not wrong. So if any of you guys as we already know have ordered an item, there's a multitude of process that comes from that item being you know customly made to actually getting front of your door

And within that step there is one thing. So the item or whatever the load gets uh shipped transported from point A to B, from point A to B to C to uh whatever the uh the end user is. And so within that system there is one fundamental flaw and that flaw is each time the cargo is actually loaded it have to be optimized. If you do not optimize the load, the chances of the item not being delivered with high quality is tremendously high. So we built this tool that uses a hybrid model. Instead of just using an LLM as an agent, we actually use a world models to actually predict the the best possible solution, the best possible load optimization per container. So uh a quick a quick view of how does it feel like? So how does uh this thing go? So what you see here is a prompt that we have actually entered and this is a uh a prediction of load within a box loading within a certain container and uh one thing that makes this pretty different is that let's say if I am a shipper let's say if I am a an Amazon FBA or some shipping company uh and I have to deliver certain amount of ships certain amount of boxes or whatever items uh the ideal placement ment of an item within that container uh depending on the box size, depending on the container size and depending on weight size have a very big effect. And for a lot of people that doesn't make sense but turns out this problem is a 12 billion problem

Uh and what the tool that we've built we actually put in a prompt basically can think of it as a cursor but for fright management and also load optimization. So let's say I uh am an admin at fried company and I have a shipment that I have to actually process and send. Uh so before the workflow goes I would go into this old software and I would import this Excel sheet and then I would plug it in and I would wait for it to actually chug it in and try to come up with some form of like predated way of setting up boxes. And turns out if that software would have worked, we wouldn't have not lost $12 billion. So clearly it doesn't work. So a better alternative is uh what we call kit. So I'm going to come here and this is a prompt that I have uh configured but uh assume we're having trouble with internet. So I may not be able to display it but as I display for the judging

So how it work is you put in a prompt here. You're going to put a prompt. Uh you explain the shipment. You explain what type of cargo that you're taking, whether if you have fragile items, whether you're uh have items that have to be delivered within specific time frame. And once you do that, you're going to send it. And while you send it, you'll have an agent that is LLM based agent prompting the world model uh to actually come up with a better way to fit the certain this box that you see the grid mathematically in the optimal form. And then it does that and then it gives the uh LLM back the answer and the LLM would make sense of it. And then we use something called Dpack 3D which actually is a a tool that is used to simulate uh those boxes that you see

And then Claude prompts that model to put those boxes in inside this frame that you see here. And once that happens uh Claude would come in and then we have a stochcastic data where we take weather data and routes. So you might ask why are we doing that? uh let's say there ha there would be some projected bad weather conditions within the shipping route. So that might have a big effect on it. So uh you might ha you might have traffic clocks, you might have certain stores or certain warehouse being closed at that time. So that might have like some form of uh problem within our shipment. So it would look at that and what you see here the route tele telemetry is it would show you a sign that there might be something at this point and you should optimize for it and it would try and self adjust. It would try and adjust the boxes adjust the dates and also give suggestions actually whether if you want to delay or postpone uh the delivery and the shipment

So to sum everything up, uh our goal is to actually help uh whether it's smallsized fried companies or big scale companies like Amazon retailers uh uh Alibaba, whether you call it Walmart to actually have a very effective uh almost zero proof supply chain fat management so that uh the incurring losses don't go to the customer nor the business. Uh thank you so much. Any questions? [applause] That's a very good question. Uh so we actually have that feature as well. We actually try to reinccorporate uh red pan to uh stream real time data uh and we have weather data and we have route traffic data. We stream the data in we prompt claw to our MCP server and then we analyze it. Uh should would this affect us? would this affect the shipment? Would this affect the delivery date? Uh and then we feedback our model and we give suggestion and also do forward prediction. I didn't actually mention that with the previous judging but we do forward prediction

So we would say based on this based on this data we might have this kind of traffic within certain time. So we might suggest a better routing or also it's if it's better even postponing or delaying delivery. Thank you. Any other question?