From Data Enthusiast to Microsoft MVP: Julie Koesmarno's Journey in AI and Enterprise Data Management

Today's guest is from Washington in the United States Well , not originally , but she'll tell you about that . She works at Microsoft as a principal product manager in Dataverse . She's also a co-author of SQL Server 2017 Machine Learning Service with R from way back in the day .

Shaping Copilot experience within Dataverse , powerapps and Copilot Studio is one of her missions . You can find links to her bio , social media , etc . In the show notes for this episode . Welcome to the show , Julie .

I do get you know friends visiting . Some are MVPs , like yourself , right ? So some of my fun memories of that having ice cream is when , like my friend Heidi actually , who is an MVP in the data platform side , we go to like shop around this like ice cream aisle and we look at this Mochi Boba ice cream and yeah , nice , nice .

So that's my favourite one right now . That's awesome .

But if I I were to visit somewhere else , I would have to say maybe Holland mostly because also for the got a few friends there really good friends there and food's pretty good yeah .

So know , some of you are probably out there that have been in the database world probably kind of knew that , but anyway , so I ended up somehow landing in the database space more , which is SQL Server or which was SQL Server at the time , yeah , and then I got awarded MVP through all the community work , the community work , um , and I figured out like hey ,

it's actually really good to , you know , be able to advocate for microsoft products and solving customers problems with the , uh , you know , with the , with the microsoft spec .

Some of you this is kind of dating me a little bit some of you probably remember of the bi burger back in the day with the ss , with the SSRS and all that kind of stuff like the data we have in Slack . So I used to kind of recall like , oh , that's kind of fun . I do wonder , like how would they do that ?

Like how would Microsoft actually think about coming up with all these different features or products ? So I decided one time just to apply and got to meet with a few Microsoft folks and yeah , and then I applied and then they let me , you know , choose which team I wanted to be part of .

And then I decided , since my background was actually analytics , I decided to join the SQL Server telemetry team . That was an amazing space to learn about how SQL Server was built , because I had to figure out all the different features that are available out there .

So , yeah , I got to work with a lot of telemetry , so it was kind of fun to work with a lot of telemetry , so it was kind of fun . And then I decided , hey , I want it to be more of a front-facing product management sort of role . So how do I get into that ?

And eventually landed in Charles' org Charles' awareness org , which is Business Applications Platform Group , and I've been very thankful of that because there's a lot more exposure for trying a lot of new things . Like you know , some of the things that you kind of mentioned during the introduction about , you know Core Pilot .

My team have been able to kind of work on that , so it's very exciting stuff .

You probably , I think you've had him on the show right ?

And so since then I was thinking about , hmm , maybe the user experience for , you know , creating tables in general or bringing data into Dataverse for especially for enterprise data or for business applications , could be way improved .

So it's just taking you know from some of my background , as you know from the pro code , but how do we introduce it so that if or when you are low-code developers , or even if you are pro-code developers , it should be pretty easy for you to create tables and bring data in ? So it all started from there .

Some of the thinking that I had at the time was can we use something like GPT ? This was before the open AI . You know big announcement on the chat GPT , so we're exploring that .

And then , sure enough , within weeks , that announcement was there and we were able to leverage a lot more on the large language models for a number of things that we are enabling for makers . So I have to say I'm very , very fortunate to be part of this journey , being able to enable for our local makers .

So some of the things that we are enabling for within the Copilot studio where you can ground your own custom Copilots with enterprise data it's cross-collaboration with a lot of different teams , of course , but one of the things that we're enabling is not just being able to allow makers to grant their custom co-pilots with data versus enterprise data , but also some

of the other what we call third-party connectors right , the co-pilot connectors which actually technically , under the hood , is leveraging the Microsoft Graph connections . So very exciting space .

So , you know , from a file share perspective , we could be on some HP trim type environment or we could be in , of course , share do have a lot of SharePoint repositories . You know that data would sit in , but it could be various SQL systems , it could be , you know , anything from POS systems to ERP systems .

When you're talking about the organization's knowledge , are you talking about all their data ?

Because day-to-day you know , individuals are not experienced queries as in . The best they could do is probably do a search engine search , and most people have used search engines way below what their capabilities are . I remember back in the day reading a book when I was in web design , and this was a very long time ago .

There was a book called Google Hacks and it was basically designed to allow you to do search querying , you know , through just the standard search tool that Google had , and , of course , you could extract all sorts of things credit card numbers , passwords , all different types of file formats , etc .

If you knew how to create a really robust query , you could access that data .

Now , with the graph the Microsoft graph coming online in the context of co-pilots , in the context of all knowledge of an organization , there's potentially a massive amount of data sloppiness and governance standards that perhaps haven't been implemented because no one's known how to find that information .

And , of course , in a world where you can , you know , ask an LLM that has reasoned over that information .

Ask an LLM that has reasoned over that information , even PII data , you know , where it might not contain the full data set of an individual , like their name , their date of birth , their address , but there might be fragments of their data all over the organization .

All of a sudden , llm , brilliant at finding patterns , can all of a sudden start bringing together a lot of information that in its individual formats would be meaningless but as a consolidation would be very interesting . Yeah , what are you seeing in this space ?

Does a co-pilot implementation really needs to carry out a security sprint following to actually do some housekeeping around the data that is now exposed ?

make sure that the data is clean enough to be consumable . So that's one aspect , but you're also adding that dimension of the security . So garbage security in , garbage security out . So definitely those are the two things that we are considering as well .

So something for , as practitioners or , as you know , customers are adopting this , we highly recommend to kind of think through how you organize your data before , of course , you know the whole . data governance is an important concept , but now it's even making it more important or prevalent to all enterprises . It's very important to kind of look into it .

Especially if you're I mean especially when you're building apps , you should look into it . But with Copilot , of course , it makes it super easy . So therefore , there's much more kind of care need to be uh given to , uh to the governance of data in itself .

And then it kind of answer like Gold Coast is in Australia . So it typically sort of like sources and provides citation hey , you know , this information comes from this . So that's what we mean by grounding with the data , so that it's specific to what the co-pilot can answer . It doesn't use like Bing today or the co-pilot that you can use from Bing .

It doesn't actually use enterprise data on your local machine . So because you don't set it up that way or nobody set it up that way .

So when you build a custom co-pilot and this is an experience that already existed even prior to build where you can build a custom co-pilot and you can say , hey , I just want to ground it or I want to use it as a data source for the code palette , just maybe only from this website or from this SharePoint site or from this file .

Just making it super easy to illustrate .

Say , if I was to upload a file that discusses onboarding guide , then if and I add it to my custom copilot so that copilot is grounded with that onboarding guide , so if I were to ask questions like where's the best sushi , then it shouldn't answer that , right , because it's just answering based on the data that I've provided it as a maker .

So when we talk about co-pilot connectors in general , it means it means knowledge , but it also means plugins , which essentially is connectors or actions , right , so yeah , yeah , and role-based access control .

You know the normal current type processes through the ui is this do those you know what you're producing therefore honor that type of role-based access ? So we don't have people , for example , say , you know , let's say , a data set in Dataverse that their role doesn't give them privilege to .

We're in what , but what's exciting you ?

Some of the things , like you know , we discussed about the data governance that will become more important . I think people are going to care more and more about their data . Yeah , lots of exciting things to look forward to .

The first one that I wanted to talk about is how , as a maker , you can get started with data modeling and inserting or creating data or creating tables rather really easily with the help of Copilot .

If you want to try this at home , and depending on when you're airing this Mark , there is a few settings that you have to do , because at the moment it's still in the early preview sort of stage . We will make it public preview very soon when you create a new environment or if you already have an environment where it's in the preview region .

So that means you have to go to the admin center and look at your environment settings to make sure that it's in the preview region . You can see this . Try the New Data Experience . Now , when you click on the Try new data experience , either you can tap in from here or start with data , or this is actually to build app .

But if you want to just stop at the table side . Then you go to the tables and then click on this beautiful button called Create new tables . Here you'll be presented what I would say as kind of like a blank canvas where you can start doing some data modeling with Copilot For the purpose of the discussion .

I'm just going to start using the start with Copilot . So essentially , using natural language , we'll start designing or modeling data using Copilot with the help of Copilot there . But you could also do similar things with SharePoint and Excel file , where you bring in your existing data to create new tables .

I'll go ahead with clicking on the start with Copilot For the purpose of this demo . I'm just going to say student accommodation reviews . So essentially , what I'm providing here is just like a prompt hey , I want to create a data model or ERD diagram , really to like a system to store student accommodation reviews .

So I can do some you know optimization in terms of like hey , do you want just the one table , or multiple tables ? What table size , et cetera , whether to include relationships or not , and then I'll click generate . So it'll take a few seconds to create .

Like , within the same experience , the same canvas , have you put ? I think some of the ones that I find interesting when you ask your pilot to do is like hotel registration systems . Usually it comes up with about three , uh , three tables or so .

So let's take a look here when you click on student accommodation , it gives you some sample data , and likewise , when you click on accommodation reviews , you get some sample data as well .

So that way , as you as a maker , as as you're creating or designing your tables , you can kind of roughly , especially if you want to create apps after you can roughly start testing out your app with some data , right , because it's usually pretty helpful to do that In case you want it to be a bit more relevant to you , in case you want it to be a bit

more relevant to you . So for me since we talked about Australia earlier , so I could ask Copilot to help me you know , change this value of CTA to CTE . It just doesn't make sense . So I'm going to say hey , change location value in student table , two cities in Australia , so fingers crossed , hopefully it works .

Again , just to want to reemphasize this is still in early preview , so hopefully it works . Ta-da , locations all changed to Sydney , adelaide , and I can ask add five more rows in student . Sometimes I'm not good in spelling , but the good news is that GPT usually picks it up pretty well . It's not like ah , you know , you have a typo here .

We roughly know what you're talking about .

So you've seen how , as a maker , I'm getting help from Core Palette to help me design my data . Now the question is what can I do with it after ? So if you're familiar with Power Apps , like Canvas apps specifically , then when you create these tables and create an app , you have the whole Copilot sidecar that you can start adding .

So that way , you can make your app to be Copilot-ready , so to speak , for your end users . Now , speaking for end users , now we're talking about how makers can create their own custom co-pilot so that their end users can start asking questions against this co-pilot .

So you can think of this custom co-pilot like as if it's a perhaps a bot that you want to add to your own website , or maybe to Teams as well . So here's one that I actually prepared earlier . So let me just make the real estate a little bit bigger here . So I've created a student accommodation finder custom code palette . It's super easy to do .

If we have time I'll show you the flow . But once you create it you can keep adding knowledge . So in this case I've already added a few different knowledge sources .

If you want to add a new one , you just click on the plus add knowledge and then you can specify public website files , sharepoint Dataverse Fabric is coming soon or any of the enterprise data that you have your graph connections set up already by your tenant admin . So just for simplicity , I'll show you Dataverse , since we love Dataverse .

Here you'll see some recommendation regarding what table should we use . Here we go . So it suggests student accommodation . The way it's suggested , it is based on the co-pilot name .

So if you give meaningful co-pilot name , your custom co-pilot , a meaningful name rather than just co-pilot one , two , three , usually the recommendation here will be a little bit more tuned to that copilot that you're creating . So I've already added this . But just for the sake of demo , let me just show you what that looks like if you were adding from scratch .

You can add multiple tables up to 15 , whether related or unrelated , it doesn't matter . Under the hood , the system or the core pilot will figure out the relationships . Interestingly , and a bit of , I guess , lifting the details a bit more here , we use what we call a natural language to SQL .

So that's our reg or Ritu for augmented generative or generated method . So under the hood , we will translate your prompt into SQL , so your natural language to a SQL statement . From that perspective , the system will just figure out what would be the relevant tables to use to answer your questions .

So once you're happy with the selection of the tables , you get to preview it , just to make sure that these are the right tables . This preview is just showing you a few columns , it won't show the entire columns . Once you're happy with it , then you click Next and then you can also provide things like synonyms .

Here , as an example , you could say let's go to the student accommodation one . Here , for example , accommodation name . Maybe you want to add things like dorm name or whatever colloquial names that are applicable to you or to your enterprise . We can also add glossary as an example , colloquial kind of names that are applicable to you or to your enterprise .

We can also add glossary . As an example , we can say hey , akomu , sorry , not accommodation . Let's say housing is the same as same as student accommodation or maybe hostel or something then you click back and then you click Add .

Once you click Add so I won't add it for now but once you click Add , it will be added to the screen here where you can also navigate to knowledge . You can edit it further . This is the one that I've created before . Previews , synonyms , glossary , it's all the same thing .

Then this is where I can start asking questions which is earlier asked , like hey , which DOM has rating of more than four ? We can ask it again . Oops , not represented . Let me just click allow . Oh , my control , see , control fee works . There you go . Yeah , so it can start asking questions and it will provide citation as well . So in this instance it will .

It will take a little bit , I think , to answer , but in this instance where , oh , there you go , it comes up with the answers already . In the instance where , let me just go back to knowledge . In the instance where , oh , there you go , it comes up with the answers already . In the instance where , let me just go back to knowledge .

In the instance where I ask questions related to who wrote a book about , you know , australia's dangerous snakes , this question is best sourced from the PDF file that I've provided , then it also provides that citation as well . So if you click on it , you get to see the preview of that , a simple preview of that PDF file .

Yeah , so this is how you test your core palette and eventually , once you do the publishing which I haven't done for this environment yet in case you want to do publishing sorry , let me just go to channels rather than publish . This is an area where you can control or you can manage the channels that you want to enable this co-pilot for .

So one of them is Microsoft Teams . So this is one that I've created earlier . So a similar co-pilot that I've published to Teams , where it actually just grounded with the accounts table , I can ask questions like where's AdventureWorks located ? It's okay in Santa Cruz .

So this is how we think the future will be , where you can now start asking co-pilot questions , not just from , for example , just from your website as a chatbot or so so , but more in the productivity tools like microsoft teams .

You know , dyslexic typing spelling is always an issue I don't have . I have a lot less issue with that now in a co-pilot world , so it's definitely an enabling type tool . Before I let you go , I wrote a post in 2018 , 2019 , and was controversial .

Um , and it's something that I like to do every now and again , I like to be controversial to get a discussion going .

It was around a lot of people were complaining around things like diversity and inclusion and things like this that you know , in my case , if you're a middle-aged white guy in tech , they don't want you to front any events anymore or anything because you're not , you know , diverse enough .

And so I wrote something with a headline to that effect and , of course , a lot of people only read headlines they don't read the blog post and , of course , all the people that read the headline agreed with me were all those people that were anti this type of sentiment . But as I go down there , as I drilled into the topic .

I said the importance of when somebody looks up to somebody presenting in tech leadership and of course , I'm only focused on tech because that's what I'm in it's important that they see somebody like me and so , therefore , if it's a teenage girl looking , they need to see female role models .

If it's any across the inclusion or diversity landscape , people need to see role models that they go . You know what . I identify with them because they're similar to me . I can see , you know , and although middle-aged white guys have had the stage for many years , it is time for change . What are your thoughts around that ? Women in tech what are your thoughts ?

So a bit of background . I do , or did have the privilege of having quite strong role models . So , for example , like my mom , mom owns a business and it's in the hardware sort of . When we talk about hardwares we're talking about like nuts and bolts , not software , software , not computer or anything like that . So it was pretty uh interesting to see .

So it's sort of like , oh , mom can do that . So I think I could probably kind of get into more on the technical side of it as well , like if needed , and there's a lot of business thinking as well as math thinking around that . So that's one aspect of me .

The other aspect of me like you know how I grew up that I also consider as privileged too , is that I did grow up in Indonesia when I was much younger and studying computer science was mandatory . Everybody got to learn it and I just happened to love it . So I usually kind of like ace the class .

So , coming to a more Western country so as an example , australia I found myself a little bit dumbfounded in in which I chose my elective , my computer science , to be my elective , and I was the only female . Also , non-english English is my second language too , yes , so I was like , oh wow , like the world is very different .

Um , so I had some of the things that kind of helped me think a little bit more that the world is just outside of me , that there are a few different sort of like diverse perspectives .

So what I from personally , from myself , I try to , I strive to give different perspectives to all , at least my perspective to other folks , and also to welcome others as well . Now there are lots of articles , discussions out there where women are struggling in this space . I'm not discounting that , definitely it's there .

So I wanted to give a bit of different perspective what it feels like to be supported , and just to a bit of a disclaimer , what it feels like to be supported , and just to a bit of a disclaimer as well . It's not like all through my career I'm always supported . That's not the case .

It's just that there are a few points in the career that in my career that I noticed oh , I know when I think I feel at my best , where I feel supported , and that looks something like this Usually the discussion within the team everybody feels like it's okay to make mistakes .

So we , and it's okay that we are on the basis of learning , we didn't know , we didn't know about something , and then we use it as a learning experience . Sometimes it's to do with technical stuff like oh , I'm sorry , I'm kind of new to this . Can you help me understand X , y and Z ?

It could be about co-pilot knowledge or whatever , or what RAG means or what LLM means , and then being able to do kind of like active listening and learning from there . But in some cases it might be a behavioral thing .

So if somebody said something a little bit , maybe a little bit hurtful , if you can and this is very , I would say , depending on the power dynamics and your situation right , if you're in a situation where you can feel safe then you could say hey , you know , can we talk a little bit ?

I think that the stuff that you mentioned before could be interpreted differently for other people , so how do we think about rephrasing that ? Also , what's the intent behind it ? Another thing that I wanted to bring up is just because it's not your intention .

So if you say hurtful things it's not your intention , though , but it doesn't mean that it's less hurt for the recipient .

We feel like we are respected , um , we can be put into a position that , hey , I'm accountable for this , my teammates are , know , making me accountable for this too , and you actually , you know , become your better self too . So those are some of the things that I've learned so far .

Where I started picking , you know , some of the behavior not behaviors , but some of the symptoms I don't know what characteristics , characteristics of when I feel a bit more supported . Um , then some other experiences . So , and I have to say , where I am right now , I happen to get to witness that a lot .

So very , very thankful of of that too yeah , that is so good .