How to Build Reliable AI Apps with OpenWebUI and aiohttp

I have written the perfect example how to PROPERLY use AI where you can rely on. This is it for now. But there are many lessons learned along the way. This information is very scarce on the internet ironically. Also, it's funny how AI models don't know how to proper AI.

Example: AI model advisor application.

This is my example project for you. It will take your input and will find the best model available that is available within your OpenWebUI account. Yes, I do have 399 models. We could say I have a problem ;). My software uses different models for many reasons. I do this mainly to keep the costs low, my software processes several millions of tokens per month. But for multi purpose, my advise is to use for most things gpt-4o-mini or gemma3:12b. I never use models advanced more than that. With some effort on correct prompts you don't need heavy models at all. Describing a whole application in a prompt to generate it, like many vibers do, is not the way. You will always end up with something you don't want or you have a very low standard. It's not about the model, it's about the prompt.

Here’s what the example app looks like:

Example Application Screenshot

Source

Source code is obfuscated to screw over low effort c/p people. I like to share some knowledge, but I do not supply production source code. That'll cost you. I have a certain life style to maintain.

See the full code here . (Yes, it's clickable. The CSS of this site is beyond terrible. But have you ever seen such structured HTML?)

.