Introduction to Pydantic AI

0:01 If you've been wondering how to build an AI agent that don't break in production, stick around and because we are about to build something seriously cool stuff uh together. So here's the thing uh we have all been there. Imagine like you are trying to build an AI agent maybe chatbot for your business or data processing assistant. you are using raw opening SDK or maybe lang chain for example like if you have uh lang chain so you might all be familiar with uh lang chain if you if anyone is doing some serious stuff so uh everything works I mean as far as the language chain Everything seems working fine until it is in the uh local system.

1:17 Uh but then you deploy it to production and s suddenly everything falls apart. So the what happens is like the responses are incon inconsistent. Your data validation is literally non-existent and debugging feels like you are trying to find a needle in a haststack. So now now if I told you this there's a framework that's designed from ground up to solve exactly these problem. What if I told you it's built by the same team uh that created Pentic uh you know the library that literally makes every major AI framework uses under the hood.

2:08 And what if I told you it makes building AI agent feels like as smooth as uh building an fast API application. So that's Pyantic AI and today we are going to build three different AI agents from scratch starting from simple and working our way up to production ready systems that could actually deploy and use in real world systems. So uh uh before anything first we'll see first you should come to the ai.pident.dev dev here you'll see the complete documentation this is what we are going to follow and so you should ask yourself this question uh when you whenever you are watching this video what is exactly identic AI if we head over to their official documentation um you can see here they describe it as a Python agent framework designed to make it less painful to build production grade application with generative AI Now the key word here is the production grade.

3:19 This isn't just another wrapper around openi API. This is a framework that's built with same philosophy as fast API. Type safety developer friendly and reliability. So think about this way what fast API did. Fast API revolutionized the web development by making it incredible easy to build robust APIs with automatic validation and documentation. The pyic AI is trying to do the same thing for AI agent. So let me show you what that makes uh things special. So if you can see the uh providers, I will head over to the example model providers.

4:06 You'll see there are a lot of uh provided OpenAI, Anthropic, Gemini, Mistral, Cohair, Bedrock, Hugging Face. So, so the models support is u really good uh works with all major providers. And the second core principle is the structured responses your like your AI outputs are validated piantic model not just a raw string and the type safety everything is properly typed so that your IED can actually help you dependency injection for testing and clean architecture and there is also the built-in observability which is the pentic AI's log fire monitoring system so you can also check log log Pentic.

4:55 This is the one where you can do the telemetry meaning you can trace all your you can have a cloud server where your open AI traces are all over present over there. So this is a paid services but alternatives are also there like mlflow traces tracing you'll we'll also make one uh more video explanation how you can use uh mlflow uh observability uh this this is all uh some serious stuff like this is a production stuff.

5:49 So now we'll start doing the uh session and we are going to start with the uh the first setting up of our