ChatGPT on your data, by Xata

hey folks it's Dave over at Zeta today I'm going to be demoing our new ask endpoints which give us chat GPT functionality to our Zeta databases uh just natively by default so that's going to allow you to do really fun stuff like right now we've used it to add a chat bot to our documentation website you can see here I'm asking it how can I create a new record using typescript and it's actually given me some sample code that would work for that use case it's also then letting me know the documents where it's getting that information from in the background this is all chat gbt it's hooked up to our Zeta database so we're querying it live in the background let's kind of show how this stuff works because you may want to do something like this for your own website so our documentation site you know we're looking here at creating records one uh here it's going to look a lot probably like your documentation site which is we just have a whole bunch of markdown files that live in a repository maybe there's some front matter on top with some keywords but for the most part it's just a mix of code and marktown that we then pass through a renderer and then output statically nothing too different pretty standard stuff there what we do a lot cooler though is on publish we then take those markdown files and we push them into a Zeta database so here we're looking at the Zeta docs database and specifically we're looking at the search table now the search table contains the slug the title the content you'd notice that the content in here is actually the markdown itself we've got some keywords there that are attached on top in fact here's that one right there which is uh the create records page now data comes with a search engine by default you can think of it as having elasticsearch sort of for free within your database that means that I can come in here and say create records and you're going to see that it's going to start filling stuff out if I wanted to I could come in and boost any of those columns so I'll come in and let's say boost the keywords column right and maybe push that one up higher you're going to see that it changed the scoring so this one specifically having keywords shot it through the roof and now it's going to return as the first result now at any time you can come through to our code Snippets area this is going to give you typescript example code it can also give you a curl response if you want so that you can come through and get the actual thing that was there so you can see for example our column keyword here is boosted this is exactly what we're doing in our documentation itself when we're just providing our regular search so for example I'm going to type that create records one and you're going to see that it's going to return the first result if we come and look at our API within the documentation system where again it's really simple we're just using that code itself so we're searching against that search table we're boosting a couple of these columns we're adding some weights on top of it and then that's what's going to give us the response now what's so different about this chatbot one where we were looking through and asking these questions for it so that's actually using a new endpoint which if you come through and look at the ask one just search for ask you're going to get the responses you're going to get a document for the API itself there that one looks very very similar to that search one because at the end of the day it's still search the difference is that we're hitting that ask endpoint and we're actually applying some rules that we're using to inform those answers that that we're trying to query so for example we don't want people asking about pricing data here uh you know how should we respond back to the user those are the rules that we're passing and then that's what's going to inform these results when we come through and ask the bot uh these various questions now if you want to play with this stuff you can come through to our GitHub repository there's a zata i o examples repository it's got a bunch of examples not just for this but also for things like next for Astro anything that you might want to get Zeta up and running with but we're looking at this chat GPT one inside of it you're going to be able to run a demo where you can sort of ping questions against a couple different databases here I'm looking at the Harry Potter one and I'm going to ask what is the spell to make you fly and apparently it's blingardium Leviosa I know that's an easy one but you can play around with it you'll note uh within this Source if you come to the zata.ts area you will see that we've got examples for the rules that we are passing to these various databases so have fun with it if you ever get lost you know where to go just go to zeta.io docs and may as well ask our chatbot how do you install the CLI it'll get you started have fun I'm Dave this was a lot of fun to build see y'all

Channel: Xata

Views: 3,548

Keywords:

Id: pbH9OHXvok0

Length: 5min 8sec (308 seconds)

Published: Wed Mar 01 2023