These experiments from @plsak demonstrate you’re running most of your services off-chain, which is to be expected, because the IC does not yet offer the infrastructure to run everything you’re providing on-chain. The vector DB might be the exception, and I am very interested to learn more about your approach.
It would be awesome if you could join the Technical Working Group DeAI meetings. We meet every Thursday, and today we had a great discussion on the most pressing needs to make on chain AI a possibility.
Your input and sharing of experiences would be very valuable for the community. 