Using an enterprise-scale vector search database might feel a bit... silly? Overkill?
But this is actually a brilliant illustration and application of some very fundamental concepts in machine learning and AI applications.
"Vectorization" is the key concept here -- turning ones' input data (whether images or sales or service records or whatever) into a numerical representation such that meaningfully similar input data is converted into numerically similar vectors.
Once you do that, all that's left is to do a nearest neighbor search, process or filter a bit, and voila -- now you're using AI and ML to build collectible card scanners or recommendation systems or heck -- even self driving cars (which is essentially what has been built here).
Overall this example is elegant in its simplicity, and useful in that it showcases that -- if you can turn your input data into a decent vector somehow, that the rest is pretty straightforward.
I've built several ML-based systems for commercial applications, and they all used vectorization + vector search in almost the exact same way as what is described here. Fantastic illustration for communicating such a powerful idea that can be applied in so many different ways.
Using an enterprise-scale vector search database might feel a bit... silly? Overkill?
But this is actually a brilliant illustration and application of some very fundamental concepts in machine learning and AI applications.
"Vectorization" is the key concept here -- turning ones' input data (whether images or sales or service records or whatever) into a numerical representation such that meaningfully similar input data is converted into numerically similar vectors.
Once you do that, all that's left is to do a nearest neighbor search, process or filter a bit, and voila -- now you're using AI and ML to build collectible card scanners or recommendation systems or heck -- even self driving cars (which is essentially what has been built here).
Overall this example is elegant in its simplicity, and useful in that it showcases that -- if you can turn your input data into a decent vector somehow, that the rest is pretty straightforward.
I've built several ML-based systems for commercial applications, and they all used vectorization + vector search in almost the exact same way as what is described here. Fantastic illustration for communicating such a powerful idea that can be applied in so many different ways.
Well done!!