New to WIP?
We're a group of makers shipping together. We help each other stay accountable and reach our goals.
Apply for accessBack
Question
Who can guide me in the correct AI direction
Hi,
So I want to try out generation AI models, thus far I tried to code it in classical if/else statements as it basically a straightforward logistics process.
But who can help me to point me in the right direction to which kind of AI model or ML model I need to use, this is the use case.
I analyze a lot of track&trace events from parcels, like:
Accepted > sorted > en route > delivered.
But there are a lot of exceptions (address wrong, returns, delays) and carriers come up with new event names and event codes now and then.
It would be nice that AI can recognize a new event code and detect if this new event code means it is delivered, it goes back to sender, or if it causes a delay.
Another things that would be nice, is if the model could predict on which date it will be delivered, based on historic data.
What kind of models are suitable for this? Can it be one big model, or do I need to create different models for each use case?
So I want to try out generation AI models, thus far I tried to code it in classical if/else statements as it basically a straightforward logistics process.
But who can help me to point me in the right direction to which kind of AI model or ML model I need to use, this is the use case.
I analyze a lot of track&trace events from parcels, like:
Accepted > sorted > en route > delivered.
But there are a lot of exceptions (address wrong, returns, delays) and carriers come up with new event names and event codes now and then.
It would be nice that AI can recognize a new event code and detect if this new event code means it is delivered, it goes back to sender, or if it causes a delay.
Another things that would be nice, is if the model could predict on which date it will be delivered, based on historic data.
What kind of models are suitable for this? Can it be one big model, or do I need to create different models for each use case?
👋 Join WIP to participate
From your questions - you need to learn the very basics of neural networks, first. Here are some fantastic videos:
www.youtube.com/results?searc…
@fafoai do you have any suggestions?
I’m more drag and drop AI, using n8n or make to automate processes.
I'm going to pretend like I just understood what you said. 🤣
Actually, it's a state machine problem. You want to draw a diagram of all states, and their transitions, and generate code from it. Take a look at github.com/StateSmith/StateSm… to get a feel.
Basically that is what I now have... I know a lot of codes, I know the logistic process in real life, I know what can go wrong... Maybe that is good enough or even better than AI/ML ever can...
90% or 95% of all parcels are okay, mainstream process. Those 5% exceptions, or parcels with issues, is where I want to be 'smart'. But to be able to train AI/ML, I need to label states to situations...And when I label them, then I already know the answer.
Maybe AI/ML is not the solution. Or maybe it is when I want to add another carrier service, to be able to detect their flow faster?
One thing you can try is to use GPT4 and see if they can label the situations themselves. Can you give a specific example of an exception that happens and what are the actions that take? If the actions are in a limited scope and the input is human readable, it might work