Generative AI app development with Cloud SQL for PostgreSQL

Key Takeaways

[Music] we're here at Cloud next24 with bala and Shambo and we're going to talk to them about generative AI development best practices with Cloud SQL for postgress sequel thank you both so much for being here with us uh before we get into it can you give a brief introduction of yourselves what you do at Google and give us a little bit of a synopsis of your talk hi Debbie nice to be here thanks for having us my name is Bala narasimhan I'm a group product manager for cloud SQL among other things I'm responsible for the product strategy and road map for cloud SQL for postgress yeah sure I'm shambu heg product manager for cloud SQL among many things I focus on making Cloud SQL the best place to build gen application for uh developers and thanks for having me here yeah so yesterday we were uh presenting our session and the focus of our presentation was uh best practices for generative AI applications when you're using Cloud SQL for postgress as a vector database we think that uh everyone today is a gen developer right every developer is a gen developer right we're in the air of democratization of gen and we think that cloud SQL for postgress is a really nice building block for that so we were talking about how to do that easily while implementing best practices amazing so can you tell me then why is cloud SQL for postgrad SQL particularly well suited for building generative AI applications like what are the advantages yeah so Cloud SQL for postgress is a man manage database service for open source postgress and and as you know uh postgress is a very popular open source database uh there's many reasons for that there's a really strong Community that's innovating around pogress and when you adopt pogress you're able to uh adopt essentially an open source database you're not getting locked into a propri database right on top of that it's really easy to convert postgress into a database for Gen applications all you need to do is execute one command uh you install an extension called PG vector and you're often running wow what a what an easy way to get started one command y great so in terms of using Cloud SQL as a vector database instead of maybe some other specialized Vector database why would we want to do that with Cloud SQL yeah so when you're building a generative AI application and you're choosing a vector database you have two you have two options there you can build you can use a vector database that's built from the ground up that does one thing and is a Vector database or you can use a general purpose database like Cloud SQL for pogus convert it into a vector database like I said with just one command and use that and we think that's the right way to go and that's because uh you know there's many benefits to converting an operational database to a vector database firstly all your data is in there already so if you had to choose a specialized Vector database you now need to move your data from one data store to another and as you know moving data is really hard and error prone uh but the most important thing really is that when you're using Cloud SQL for postgress you're getting all the Enterprise features with it you're getting availability out of the box you're getting data protection out of the box but most importantly all the security and gance requirements around data is taken care of because you're just you're using Cloud SQL for postgress and Cloud SQL is ready for that and in contrast if you're going to use a specialized database you need to make sure it checks all those boxes so Cloud SQL could be a vector database right you don't need to go specialize that's wonderful so if I am a user and I want to get started how can I do that yeah sure so because of gen there are so many new possibilities and everyone is thinking of building gen application so when someone builds a gen application time to Market is very important you don't want to spend several months or years building a solution you want to ship something fast and then iterate on that so that's why uh we have published the jumpstar solution uh built using gke Cloud SQL for post SQL and vertex AI using this jump start solution anyone can build a gen application in less than 30 minutes so even if someone doesn't have AIML expertise it is possible to build gen application in less than 30 minutes using this jump start solution so we have published this jump start solution in uh uh GitHub it contains all the codes and templates terraform templates Etc within few clicks someone can actually get started and they can get started right today so uh we are providing the link in the description so someone can get started with this so what this jump start solution exactly does is you know it gives the data for you it you know uh creates the vector embeddings then it also creates a gen powered chatbot that can actually interact with users and help them find the right products so I really encourage everyone to give it a try and start building J application from today itself amazing yes I love jump Solutions we've recently been deploying a lot of them and there's such an easy way as you said it really lowers the barrier to entry and it gives you kind of that boiler parade of how can I get started quickly and then you can edit from there as you need to for your business that's wonderful so you've deployed this jump start solution what are some best practices to keep in mind what is it important for people to follow so that they're successful yeah so from application development perspective when someone deploys a gen application in production the observability becomes very important right so so you have to make sure that the performance is good query performance is good everything is working expected Etc so that is why Cloud SQL for post SQL has provided geni specific observability so that it is easy for anyone to see what's happening in the database what's happening in the application and then they can optimize based on that uh Cloud SQL obviously provides all the system metrics whether it is CPU utilization dis utilization Etc but apart from that specifically for Gen when you are doing a similarity search based on Vector embeddings you really want to make sure that the performance is good so because of that cloud SQL provides a data cache which helps in speeding up those queries but how do you know whether data cach is being used efficiently or not so that's why C Cloud SQL offers different metrics like how much you know data cache is used what is the hit count miscount all these metrics are available for customers along with the product itself so it's very easy to observe what is happening so it's not just about metrics you also want to dive deep into the query specific itself right like what is the query doing like is there any bottleneck Etc so for that cloud SQL has something called as like query Insight so it is a tool inside the product itself so you can go to Google Cloud console and look at query insight for in-depth observability especially it provides you visual query plan so for anyone it is easy to see what is happening inside the database and uh further optimize the uh Performance Based on that and as a Next Step uh they can also do end to end tracing from application to database when they use Cloud SQL because Cloud SQL supports such tracing abilities so it is very easy for someone to get started and also do the observability in production yes that's amazing so essentially using cloudsql having the metrics in there and the service the features in there that it has you can follow those best practices and you'll be good wonderful thank you so much for being here with us today I really appreciate it make sure to check out the links in the description to learn more about using Cloud SQL for postgrad SQL to build generative AI apps and uh have a great great rest of your next yeah thank you [Music] [Music]