100% Reliable LLM Outputs with Structured Data (Developer Tutorial)

[@rickhoro]

Nearing the end of the 3rd video in this series I can say that this is by far the best information I've seen on Pydantic AI. A great tool and a great video series. Thank you!

[@AISoftwareDevelopers]

@@rickhoro thank you for the kind words. I am glad you enjoyed the series. Stay tuned for more chapters- graphs are exciting!

[@contractorwolf]

love the way that this guy explains everything. Very useful, very easy to understand!

[@AISoftwareDevelopers]

Thank you for the comment. I am glad you find the content easy to understand.

[@iwswordpress]

Great to see real world uses for PydanticAI. Looks like one of the best series on KZbin for this.

[@AISoftwareDevelopers]

@@iwswordpress thank you for the comment. I am glad you enjoyed the video 👍😁

[@jonm691]

Thanks for the detail and the range of examples. Loved how reusable this pattern is across so many solutions

[@AISoftwareDevelopers]

@@jonm691 thank you taking the time to provide feedback. It’s viewers like you that provide creators a direction, and that’s super valuable in a world of constant change.

[@gr8tbigtreehugger]

Another super helpful video! Many thanks!

[@AISoftwareDevelopers]

Thanks for the love. I'll go out and hug a tree in exchange ;)

[@pritshupanda7262]

A good use case for this would be parsing personal bank statements and generating a monthly budget/expenses report. Add in agents that generate charts and graphs. Can also add agents that suggest tips on reducing monthly expenses by identifying unnecessary expenses. Finally export as pdf. Automate the process and you can have a monthly budget/expenses newsletter in your mailbox.

[@AISoftwareDevelopers]

That is an excellent idea. Let's make it happen! The key will be to get to the information, so this will have to be run by each person or in a cloud environment where they’d upload the statements. How would you run this? 😁👍

[@A.n.i.N.o.o.k]

Great video! 👍

[@AISoftwareDevelopers]

I’m glad you found it helpful. Cheers!

[@AISoftwareDevelopers]

Thank you all so much for chiming-in ❤😊🎉. The video has generated much interest and interesting ideas are pouring in. Please continue to share your thoughts for possible use cases in parsing data and using structured documents. It seems the possibilities are endless. 🤓

[@habahrami]

Perfect, thank you

[@AISoftwareDevelopers]

You are most welcome! Glad you found it useful. Cheers to a great 2025! 🎄🎁🚀

[@mahajanravish]

try moondream model for OCR

[@AISoftwareDevelopers]

Roger that, thanks for the suggestion.

[@dreamphoenix]

Thank you.

[@AISoftwareDevelopers]

@@dreamphoenix much appreciated 👍

[@pavanyadavalli6888]

will using deepseek r1 instead of the 4omini works same? i had a doubt on this ...

[@AISoftwareDevelopers]

No, they wonk work the same. You can’t use tools or ask for structured outputs from R1. Yet! 😆

[@pavanyadavalli6888]

@@AISoftwareDevelopers exactly what i thought , thanks and what open source model you suggest to use and gives structured output on par with 4omini

[@AISoftwareDevelopers]

@ llama-3.1:8B or mistral-nemo both should work just fine

[@Jobeyhshxgs]

Give Anything LLM a try for local LLM. It really is the best, so easy to use.

[@AISoftwareDevelopers]

It's in my queue already! Thanks for the suggestion 💯

[@wangbei9]

Just feel without pydantic-ai, we could do the same thing with OpenAI library or having instructor as a wrapper on top

[@AISoftwareDevelopers]

Roger that. Without PydanticAI there are a number of ways to accomplish the same thing. LangChain/Graph also provides ways to do it and as you metioned, if you go straight to OpenAI, it's rather easy. The difference is that it is built-in with PydanticAI and it makes it convenient to use, therefore it removes the friction. Thanks for the comment 👌

[@Jobeyhshxgs]

I also think pydantic is so much better for production so easy to swap out llms based on requirements.

[@AISoftwareDevelopers]

Yupp, it makes sense to add a layer of abstraction before hitting an LLM. Models change, prices go up, companies go bust...frameworks like Pydantic give you that little cushion of protection. Thanks for chiming in.

[@The-ism-of-isms]

Can achieve same thing in flowise ?

[@AISoftwareDevelopers]

Absolutely yes! 🙌

[@pabloescobar2738]

I think this is a big problem all people, you say force all person buy nvidia its monopoly😅🤫😉

[@AISoftwareDevelopers]

You know you can run all the examples in this masterclass without a GPU…no need for some expensive piece of hardware 👍

[@pabloescobar2738]

@AISoftwareDevelopers i have amd 6700amd 12gb vram at the moment 1 model run 🫤, i need Nvidia if i want run more models😉

[@AISoftwareDevelopers]

That's a great machine you have already, but yes, for more serious AI work, a GPU-rich setup would be ideal. I...on the other hand...don't even have a GPU, lol. An entry-level MBP with M3/16GB RAM is enough for most 7B models. For anything larger, I go to the cloud.

[@pabloescobar2738]

@@AISoftwareDevelopers if you want run stable model 3b/7b you need 24gbvram, if you run on cpu minimun 32gb RAM. Its similar medium. I have raison went i said nvidia monopoly😉

[@AISoftwareDevelopers]

Roger that.