Talk To Your Database - A GPT Multi Agent Postgres Data Analytics Tool

the way we code is changing and so is the way we interact with our databases every product starts with a simple question we started this series by asking what if you could talk to your database in natural language without burning your precious time writing debugging and reading SQL statements this is one of many questions asked over our 10 video postgress data analytics series that ends today the quality of our outcomes is determined by the quality of our questions and in each video throughout the series we ask different questions to push our abilities as AI charged product Builders into the future what if we can't solve a problem in one prompt how can we control our multi-agent systems how can we reduce costs and monitor our agentic software can we access more memory while managing less State can our AI agents correct their own mistakes with each question came a new video where we pushed to find new answers all of our work has led us to this final video with our answer to our original question from video one in the form of a product I want to introduce you and welcome you to talk to your database a new way to interact with your data built exclusively for data Engineers analysts product Builders new products typically offer one or more of these three value propositions better faster cheaper talk to your database aims to be better and faster at helping you communicate with your database let's see how talk to your database accomplishes that talk to your database is launching exclusively right before your eyes to join the BEDA launch type [Music] 9999 by watching this video you are exclusively part of the launch this product is our secret we'll update the passcode every week until 2024 to keep the product as as lowkey as possible we want to keep the product really lowkey as it's in beta and make improvements up till 2024 for the final release create your account here I'm going to toggle to log in and let's get started after you log in or create a new account we're greeted by hyper minimal UI with a couple components we have an input field a list of query results navbar and a sidebar tab with our table talk to your database automatically connects to a demo read only postgress database this allows you to test with zero commitment zero risk and minimal setup now let's look at the core of the product the natural language input field if we hit t or click the show tables tab we can see all the tables in our postgress database we can see here in the users table we have this off booing I'm going to type select all off users you can see here the table context popped right up above the input field and now I'm going to hit enter you can see here select all off users popped into our query results and is loaded if you've been watching the series you know exactly what this is doing in the background awesome our query has Auto updated and has presented several options to us now we can view the results we can copy the SQL or we can download the results let's go ahead and just click the row and see what we get we have our natural language query we have the status we have the table context which is the tables that our AI agents use to complete the query you can see here the generated SQL query and our final results in a nice table that we can sort and view the results of all users are offed one of my golden principles of engineering never trust new technology validate the results to build trust so let's go ahead and verify that these are actually the correct results if we open up table plus let's copy the SQL paste it in and run it as is we have about 51 rows and if we just sort by ID we can see the first 10 and if we look back over sort by ID here you can see roughly the first 10 results are exactly the same with talk to your database you can be incredibly short and concise let's create another natural language query let's ask users join job limit 10 let's go ahead and run that you can see we have a brand new query that's loading and I want to just highlight again how short and concise this natural language query is it would be really hard to express this query in fewer words users join job limit 10 so for click results here we can see we have a brand new SQL statement we can see our table context consists of the jobs and the users table we have the users we have their respective job and we have a limit of 10 so again you can just copy this out paste in these are the exact results we have in talk to database so this is awesome I want to add some more clarity around the column results so let's go ahead and just ask talk to database to do this for us I'm going to reuse this query by clicking on the natural language query in the row and then I'm going to add prefix column names with respective table let's go ahead and let that run this is something we can add automatically in the future as a customization setting but for now we're just going to type it in natural language and get the exact results we want awesome so as you can see here we have a new select statement with the prefixed results we can see exactly what table each column belongs to often times you'll have tables that both have updated columns or ID columns so this can be particularly useful then so let's put our SQL AI agents to work let's run another query open up our table cbar and let's look for our notification table here we go we have user notifications let's say I'm doing some data analytics and I want to know how many user notifications have not been read I'm not going to type SQL I'm not going to do any of that I'm going to open up talk to your database I'm going to come to the input field and just type that exactly how many user notification haven't been read yet let's view the results we can see here aund users have not been read yet again we can copy the SQL and check the results look good we can take this query push it further figure out which users aren't opening their notifications for whatever reason the beauty of a tool like this is that it allows us to build on our existing queries we can get results and then continue our exploration in natural language and at least gives a nice trail of history of everything we've been asking and all of our results so now that we know a 100 user notifications haven't been read let's see which users aren't reading their notifications so which users are not reading their user notifications you can see here I'm making sure that this table gets in our AI agent table context by explicitly stating the table there are ways to work around this and we can lean on our AI agents to do this for us but we'll always get better results if we just explicitly list the table all right let's check the result awesome so you can see here we're getting the user ID name and email where they have unread notifications awesome at any point we can just copy the SQL and at any point we can just download the results so far so great right we're talking in natural language and getting concrete results and SQL out of our database even Pro database admins would be saving time now let's be real right we've all seen product launches and product demos where presenters are basically just cherry-picking the happy path of their application I'll be the first to say talk to your database isn't perfect and I don't want to paint it that way let's push the app to its breaking point if we open our tables search for products we can see here we have products and supplier products right now what I want to say here is something a little ambiguous right let's say select the most popular product okay so just by looking at the table definition here you can see that there's no real way to determine if a product is popular we still got a result back it's doing its best guess by saying you know it's ordering the stock ascending limit one so it's basically saying the fewer products you have the better the product is likely going to be right and this makes sense but as the domain expert I know that that's not exactly how we measure how great a product is we're actually looking for the product reviews table to tell us how popular it product is we want to know not the lowest quantity we don't want to know the price or the price and the quality what we're looking for is the product that has the best reviews let's go ahead and just look at some data in this table list 10 random product reviews talk to database is also just a great way to quickly get familiar with different tables so you can just quickly ask tell me what the data looks like give me a couple random results so we can see in this data every product has a a rating here and what we want to do to find the most popular product is grab the product that has the summed highest rating so we want to grab the product that has allog together the highest rating let's go ahead and type that out in natural language select the most popular product based on the sum of its product reviews rating so once again you can see we have the products and the product reviews tables that are going to be sent to our AI agents to figure out what query to run by having them here and EXP licitly mentioning them in your prompt you can be 100% certain that they'll be a part of the query let's let that run and let's see what we get all right let's check our results and let's see what we have here so select product sum product rating and then you can see here this is being aliased and we're aliasing it as total rating and then we're sorting descending this is great and then we're going to do a limit one that's going to give us the top product with the rating and you can see here product 75 has the highest com bind rating this gives us our most popular product we got these results all in natural language just by thinking about the problem at a high level as you would before minus all the work effort and details that go into actually creating running debugging your SQL statement so every example we've looked at so far has been relatively simple let's look at a really interesting query that Nails the value proposition of talk to your database in stone let's look at the order table and I'm going to ask a deceptively complex query for talk to database to answer you can see we have this order date field here I'm going to ask average time between orders that's it four words let's let that run if you're a database admin data analyst or full stack engineer you know that working with time and finding distances between time in SQL is not simple it's not a simple walk in the park there's a lot that goes behind this let's open up and see the results of this deceptively simple query and yeah exactly we're building out our ordered orders based on date there's some lead and order by statements we have this Over clause we then have to determine the distance between each statement and then we need to make sure that those statements aren't null and then finally we get the average this is not a simple statement but as you can see we typed four words and generated a ton of SQL that gave us a result the average time between orders is 3 days and then it looks like 14 hours 32 minutes Etc I just want to highlight this point I'm going to go ahead and open up the console let me just copy this let's get the length of this right so we tyed 27 characters let's copy the SQL let's paste this in and we generated 390 characters of SQL if we divide this we can see what our multiplication rate was how much faster were we talk to database helped me write that statement 14 times faster it's actually a lot more than 14 times because we're assuming that you and I are a datab Bas admin Elite Guru who can you know type this SQL statement in one shot at the exact same rate that we can type in natural language which is just not true even for the most advanced SQL admin that's a 14 times multiplication rate you know I want you to close your eyes for a moment imagine you become an order of magnitude times faster than anyone you know at one of your engineering activities literally 10x faster now I want you to open your eyes and talk to your database enables you to 10x your communication with your database so let's talk about security if we click connect you can see we're using a demo connection with a demo readon postgress user and a demo open AI API key whenever you're ready whenever you feel comfortable you can disconnect and use your own personal connection I highly recommend you use a readon role a readon user I'm going to go ahead and do that right now once you've entered your information go ahead and hit connect and as long as you're not behind some type of VPN or something like that you should should connect successfully so if you're enjoying your experience using the demo database connection and you want to try your own connection here this is the place to do it so we've seen data hacks time and time again I'm not interested in them you're not interested in them talk to your database automatically nukes your data big tech companies want your data but me and my team of AI agents don't want it if you click customize you can see here we have retention policy and connection timeout retention policy is the amount of time that your query results will remain on the talk to database servers after 1 hour your query results will automatically expire connection timeout is the exact same thing for your database connection you can stretch this out in the setting to 5 days at the longest only you are the one with true access to your database and true access to the results of your database in additional layer security that talk database has we have the notion of stop words so anytime one of our AI agents generates an SQL statement that contains a stop word it will automatically reject the statement let's go ahead and look at that right now let's try to insert some data right insert a new row into the users table so I'm going to hit enter and our multi-layered multi-agent data analytics AI data team is going to notice that we have a insert statement here let's reject that statement completely so you can see here we have an error and it's reporting that it found the stop word insert so if we click in here there's no information there's nothing to find there's nothing to see because the query was inv validated we found the stop word and so this is what the feedback system looks like in talk to your database every error is going to show you some type of result it's going to tell you why didn't it work right this adds an additional layer of security again on top of your readon connection that's not the only set of settings here you can scroll and see all the different ways you can customize talk to dat datase to fit your exact style you can see here we're only showing the previous 10 rows we can hide and show actions we have this automatically copi a clipboard functionality and much more coming in the future of talk to your database so just another note on security I completely understand if you don't trust me if you don't trust o database this is where a read only connection is really important you know technology enables us to not trust each other while still getting along fine there's probably some darker societal repercussions to that but we'll save that for 2024 predictions video of course whenever you want you can log out and the system will automatically expire both your connection and your career results so I talked about expiring your results let me just show you exactly what that looks like I've got an edge pro version of Talk to database on my monitor here I'm going to go ahead and run a query that's going to automatically expire one of these queries so you can see what that looks like so you can see here our select most popular query has been expired it has this different grade look it's got the timeout you can see here there's nothing to download now we can still copy the SQL but if we click in here you can see we just have this elect statement it's not leaking any actual data this row has been expired over time you'll always see your queries become expired this is just a matter of data security again big tech companies want your data I don't want it talk to your database doesn't want it it's more management it's more headache it's more risk so we get rid of it talk to your database keeps your data safe by not having it at all so there's one thing we haven't talked about here talk to your database enables you to be a part of the development I want to keep the feedback loop between you and I really tight so all the product feedback all the development rope map bugs and questions will live right in the product you can see here I have some mock feedback feature requests and some questions right now you can create feedback of three different types bugs feature requests and questions whenever you leave some feedback you really help improve the product based on volume I know I won't be able to respond or get to every issue but this will help improve the product rapidly so feel free to drop in feature requests this is a Dev heavy product so your feedback is super important no product launches perfectly no product is the right version when it's first released when questions bugs or new feature requests start moving into motion they'll automatically get updated here and you'll be able to watch it move to completion and get the feedback and get responses to your questions and ideas here try not to be spammy or annoying or angry or anything like that the talk to your database AI agents will Mark your account as annoying and probably ignore your feedback on the flip side if your feedback questions and feature improvements are legit and help move the product forward the moderation agents are going to promote your feedback to the top of the stack we're just scratching the surface of everything talk to your database can do and everything talk to your database will be able to do with your feedback with some time this product has become the new way that I interact with postgress databases I'm actively using talku database to build out your database and it's really turned into this really positive feedback loop and I think you'll find it useful as well especially if you interact with SQL on a daily basis there's a lot more going on here but really the core value is this you now have the ability to talk to your database in natural language you know like I mentioned it's not perfect you can throw tons of things at this you know just random crap that will not give you valid results and there's still some work that needs to be done to fine-tune the product so that it's automatically understanding your data and giving you back exactly what you need but as a product builds we'll be able to tweak modify and optimize the AI agents acting behind the hood to generate the exact results you want in the fewest number of words the biggest issue I'm running into right now is speed this product is built on top of GPT technology GPT 4 specifically by open AI as anyone who's been following the channel knows the big problem with that technology is that it adds an additional API request and as product Engineers know every API request you add to your product slows it down so we'll be looking for ways to maximize speed within the application within the bounds of what we actually can control but that leads us perfectly to where we're headed next you know from the beginning I wanted this channel to be a place where product Builders can gather learn exchange ideas and really build products we built a functioning product in public we shared and developed ideas around the latest and greatest Tech like gp4 assistance API autog gy and more the channel has really taken off and I just want to thank everyone who's been a part of the journey especially if you're an OG that joined before the blow up but I also want to appreciate everyone new that's joining the journey now there's so many channels that just Yap about the latest AI hype cycle news and build toy apps right from the documentation we're here really exploring truly experimenting and taking products from zero to one you saw it here with the postgress data analytics tool it is now something you can use something you can feel feel something you can appreciate something you can learn from right talk to your database is live and it's the product of all of our work over the previous 10 videos on this channel we take products from zero to one that is what we do if that sounds like you if you want to be a part of that if you want to do that yourself really complete things take things all the way create value hit the sub and join the journey 2023 is just the beginning so you know while we're chatting let's just go ahead and just fire off a couple uh you know old queries whenever you want you can just hop in select some of your old queries fire them off this is just kind of cool I love seeing multiple queries running at the same time if you know several things you want to do you can queue them up and it's a really cool way to you know get a lot done in a short amount of time so you can see here we have three queries running actively and the results will flow in one by one we can see we got an error here for some reason it found a stop word so we can just go ahead and rerun that no problem and you can see here we have additional results so what's next for us the postgress data analytics series is over from it you and I have a new way to interact with your database as many product Builders subscribed to the channel know the launch of your product is just the beginning here's what you can expect in the future of talk to your database we want to build out a clean autocomplete we want our SQL agents to be able to self-improve we want to add voice commands we want to add favorites we want to improve the road map and we want to improve the feedback we're also going to have a pro version coming out with some pretty insane features in 2024 recursive data generation safe insert update and delete functionality super mode and exclusive dedicated AI agents to speed up the process so for everyone supporting the channel and for all the product Builders who really see the value in 10x in your database interactions talk to your database as a 50% or early sign up discount up until the new year so if you sign up anytime between you're seeing this video and up until 2024 you'll get 50% off the bleeding edge pro version that will launch with the New Year in 2024 so you know I'm going to be honest I'm having kind of a hard time wrapping up this video um I really get sad like this but uh you know the series is ending and uh as soon as I turn the camera off our postgress data analytic series is officially over it's been a it's been a really great journey and I'm glad that you've been here along the way you know one one of my biggest Inspirations is knowing that there's someone out there watching this channel that's going to take what we're doing here all the work we've done leading up to this video all the value that's in this product all the value that we're going to create in 202 for someone's watching this Channel and they're going to take all this to the next level and and that just really gets me excited you know to be able to contribute like that and to be able to help other engineers get to that next level not only to stay relevant in the age of AI but to be on the edge to be on the bleeding edge of what's possible this is just the beginning of our journey to become agentic engineers so one last thing here with the series ending it's probably going to be a good time for me to step back a little bit focus on talk to your database focus on building a great application and also plan out lock in 2024 we have a lot of exciting things coming but I have to say you know I have a new found respect for YouTubers for people creating content keeping up this pace every single week even every other week um is a pretty massive task and you know it might seem like it might seem like things are happening magically but there's a lot of effort and work that goes into creating developer content especially when you're actually building something you're actually you know trying to share the the core of ideas and different structures and use different tools it takes a lot of time to really think things through in a way that is productive for someone to you know watch and consume so it's probably a good time for me to step away for a little bit take some personal time reset recharge to blast off into a brand new year while focusing in on talk to your database I want to give that the attention it deserves and I want to make sure that what we've built matters I want to make sure that what we've built can really create value so last note on that again huge thanks to everyone who's joined the journey if you're watching if you made it to the end um and you're you know you're really excited about htic software you're excited about talk to your database whatever it might be you know you're looking to build your own product hit the like hit the sub join the journey if you're interested check out the product totally free to sign up there's zero risk and trying the product you can get out whenever you want or stay in as long as you want I just want to say thank you it's been an incredible series The Channel's really taken off and as I've said before this is just the beginning 2024 is going to be wild I'll see you in the next one

Channel: IndyDevDan

Views: 3,777

Keywords: ai agent, ai agents, ai data analytics, multi agent, postgres ai, ai sql, talk to your database, ttydb, autogen, gpt, gpt4, turbo4, assistants api, agentic software, autonomous data analytics, ai bi

Id: 5wROK4lBoeo

Length: 24min 45sec (1485 seconds)

Published: Mon Dec 04 2023