[English] Heli Helskyaho @ OUGN 2023 in Oslo - Machine Learning and Family Business

00:00:00: [MUSIC PLAYING]

00:00:03: Hello, and welcome to another episode of Devs on Tape,

00:00:15: this time from Norway, from the OUGN conference in Oslo,

00:00:21: ladies and gentlemen.

00:00:23: On today's podcast, we are thrilled to welcome

00:00:25: a true pioneer in the world of technology.

00:00:27: She's already laughing.

00:00:29: She is an accomplished entrepreneur and international speaker

00:00:32: and a recognized expert in the field of database management.

00:00:35: With over 25 years of experience in the industry,

00:00:39: she has established herself as one of the most respected voices

00:00:42: in the world of technology.

00:00:44: As the CEO of Miracle Finland OI,

00:00:47: a company that specializes in database management

00:00:49: and data integration solutions,

00:00:51: she has been the forefront of digital transformation

00:00:54: that is sweeping across the world.

00:00:56: She has helped countless organizations

00:00:59: leverage the power of data to drive innovation and growth.

00:01:02: And her insights have been sought after

00:01:05: by businesses of all sizes.

00:01:08: She is also an active member of the global technology

00:01:11: community, serving as the president of the Finnish Oracle

00:01:13: User Group and a board member of the International Oracle User

00:01:16: Group.

00:01:17: She has been recognized with numerous awards

00:01:19: of her contributions in the industry,

00:01:21: including the prestigious Oracle Ace Director Award.

00:01:25: Today, we are honored to have she join us to the podcast

00:01:29: to share her knowledge and insights

00:01:31: on the latest trends in technology.

00:01:33: She challenges the facing the industry

00:01:36: and the opportunities that lie ahead.

00:01:38: So without further ado, let's dive in and welcome Heli,

00:01:42: oh, that's hard, Hels Ki-Yaho.

00:01:45: - Almost. (laughs)

00:01:46: - To the show.

00:01:47: Hello, Heli. - Hi.

00:01:49: - We stick to Heli, right?

00:01:50: - Yes, Heli from Finland.

00:01:51: - Heli from Finland, that's the way we can work with that.

00:01:55: All right, yeah, I'm very thrilled to talk to you today in the podcast episode.

00:01:59: And yes, unfortunately, Caro didn't make it to Oslo, so we are just the two of us.

00:02:05: And we have plenty of things to talk about.

00:02:08: So maybe you can just introduce yourself besides my introduction to you and tell us

00:02:13: everything, where it starts, where did you start going into the technology branch and

00:02:18: everything about maybe your company, Myrica Finland, or why?

00:02:23: Well, first of all, I think it's a surprise to everybody that I am on IT because that

00:02:28: was not the plan.

00:02:30: I never thought I would be on IT.

00:02:32: It was not something that I even considered.

00:02:35: So I was actually studying mathematics.

00:02:36: I was thinking I will study mathematics or economics.

00:02:39: So I'm still actually studying economics.

00:02:40: I haven't graduated from there yet, but I was studying mathematics and then I took some

00:02:44: minor courses on computer science and they made no sense to me.

00:02:47: I was like, come on, this can't be true because it felt so hard.

00:02:51: Studying has always been easy to me.

00:02:53: I always loved studying.

00:02:55: I was like, this is so hard.

00:02:56: How can it be?

00:02:57: And I took another course and another course

00:02:59: and another course until I finished all the minor courses.

00:03:02: And then I was thinking, okay,

00:03:03: I still don't understand this computer science.

00:03:05: So I changed my major to computer science.

00:03:08: Well, it was a funny conversation with a professor

00:03:10: from mathematics because he was like,

00:03:12: "Heli, are you sure?

00:03:13: Because you are good in mathematics,

00:03:15: but you are no good in computer science."

00:03:17: - Yeah, just in the courses, yeah.

00:03:19: you still want to change to computer science? I said, Yes, I do. Because I want to understand

00:03:24: there must be something in computer science. So I can't buy this. It can't be this stupid.

00:03:29: It has to be something that I don't understand yet. Yeah. And yeah, it was because databases

00:03:34: was my thing. Until that it was like operating systems, networks. Oh my god, I still don't

00:03:38: like them. Yeah, sure. The basic stuff, right? Yeah. But databases and data, that was something

00:03:44: that I was like, Oh, now I got it. I got it. This is this is what I want to do. So I graduated

00:03:49: with a master's degree from computer science and that's it.

00:03:53: And then you just opened a new company, right?

00:03:56: Yeah.

00:03:57: Directly or was there some time between?

00:04:00: I was working for Oracle for a while, by the way, before this.

00:04:04: Oh, yeah.

00:04:05: I didn't get this, no.

00:04:06: Yeah.

00:04:07: So from Oracle, I started my own company in 2000 and that was called Kandamestarit.

00:04:12: It means like base masters or database masters, something like that.

00:04:16: We started that in 2000 and I worked for that for quite many years.

00:04:19: I actually still have that company, but then I joined Miracle in 2010.

00:04:23: Okay.

00:04:24: So, and then the time starts when you were presenting at conferences, right?

00:04:29: So, I mean, almost every schedule of a conference I see "Heli from Finland".

00:04:33: Yeah.

00:04:33: Just avoiding to say again.

00:04:35: That's what everybody says.

00:04:37: Yeah.

00:04:38: Actually, I started in 2007.

00:04:39: So, when I started to be active in EOUC, the EMEA user community,

00:04:44: So then I started speaking.

00:04:46: I went, my first presentation was in Croatia,

00:04:48: the Croatian music group.

00:04:50: And that's how it all started.

00:04:51: I realized I really liked this.

00:04:52: So I have been teaching a lot.

00:04:54: So throughout my career, I have always been teaching.

00:04:56: So I kind of like teaching,

00:04:58: but I didn't really present in conferences until that.

00:05:02: So 2007, and that's how it all started.

00:05:05: Suddenly I was everywhere presenting

00:05:07: because it was so much fun.

00:05:09: - I was just thinking about what I did on 2007.

00:05:12: I don't think anything near IT stuff, right?

00:05:15: (both laughing)

00:05:17: And now you ended up writing numerous books, right?

00:05:19: So my opinion on that is a little bit different,

00:05:24: I think, because writing books in the IT

00:05:26: is like writing text, getting everything in place,

00:05:30: publish it, like print it, publish it,

00:05:32: and then sell it and everyone reads.

00:05:34: And at the time of being, when this is ready

00:05:37: and laying on my table, it's already outdated, right?

00:05:39: What are your opinions about that?

00:05:40: So you're grinning, right? - Yeah, that's true.

00:05:42: That's true.

00:05:43: But my first topic was about database designing.

00:05:45: That's never old.

00:05:46: So it's always the same.

00:05:47: Okay.

00:05:48: It also talks about data modeler, which has changed quite a lot since 2015 when it came

00:05:52: out, but the book came out 2015.

00:05:55: But it hasn't changed that much because the designing work is still the same.

00:06:01: So I think that's still valid.

00:06:03: Still valid.

00:06:04: And the second one is about SQL and PL/SQL.

00:06:06: Where would that go?

00:06:07: There's a lot of new features coming, but the old features are still there.

00:06:11: So it's still okay.

00:06:13: Then the two other books are about machine learning.

00:06:16: The first one about machine learning to Oracle professionals.

00:06:19: So that introduces how Oracle is handling machine learning.

00:06:22: So what kind of tools you have and how to use them and so on.

00:06:26: Well, that's still valid, but there's much more tools available now than they were at

00:06:29: the time.

00:06:30: Yeah, sure.

00:06:31: And then the latest one with Adrian Ping is about Apex.

00:06:35: Oh yeah, Apex is never old too, right?

00:06:37: It's never old.

00:06:38: Apex is young forever.

00:06:40: Now, seriously, that's interesting because we are talking about APEX, OCI, machine learning,

00:06:46: all that kind of things.

00:06:47: So, as an APEX developer, what should you know about OCI?

00:06:50: What should you know about machine learning and so on?

00:06:52: So, I don't think it will be old very soon.

00:06:54: Yeah.

00:06:55: I mean, I have my own history with that because there was Jürgen Ziem writing a book about

00:07:00: APEX, like the big compendium of everything with APEX.

00:07:04: And I was asked to write the chapter about mobile.

00:07:08: And it was right before the Universal Theme was released and there were rumors and saying,

00:07:12: "Okay, you have one theme for everything."

00:07:14: It was responsible.

00:07:15: And I just wrote like pages about the mobile theme, jQuery mobile stuff back then.

00:07:21: And at the time when the book was released, the Universal Theme with Apex 5 was there

00:07:25: and it was outdated.

00:07:27: Just the passage where I was saying, "I guess there might be a theme which is completely

00:07:33: fully responsive to smaller screens and there might be something cooler than

00:07:38: what you're reading right now and it happens, right? So this is my opinion

00:07:42: that's the reason why I'm never... I have similar things there as well that are

00:07:46: already outdated but... I mean you can update that, you can have a sequel of the

00:07:49: book and you can talk about the newer stuff and if you switch over to

00:07:53: universal theme for example you might have luck for the next couple of

00:07:57: years that your book is still up to date, right? But are you just writing books or

00:08:03: Or do you write blog posts or like online media?

00:08:08: I used to.

00:08:09: I have a blog, but I haven't had any time to write lately.

00:08:11: So I'm always planning I will start writing again, but then I'm too busy with everything

00:08:15: else.

00:08:16: Yeah, you have to be continuous working on that blog, right?

00:08:18: So that's the difference.

00:08:19: Exactly.

00:08:20: So if you're writing a book, you're just having your time whenever you have it, right?

00:08:24: And then write your texts and chapters and every...

00:08:28: At some time the book is complete and you can publish that.

00:08:31: If you open a new blog and you're just writing a new article and then a few months nothing,

00:08:36: then it's not a good blog, right?

00:08:37: So we are on the same page with that.

00:08:40: So you already said machine learning.

00:08:43: I will not out myself as not into that topic.

00:08:47: Maybe you can use easy words for me and for our listeners, maybe for people who are not

00:08:52: familiar with machine learning.

00:08:54: What's the essence of machine learning?

00:08:56: How far are we already?

00:08:58: And how should I start learning about machine learning if I want to?

00:09:02: Yeah, the question of how far is a difficult question, because it depends on how you measure

00:09:07: it.

00:09:08: So there's a lot of cool things happening at the moment at the area.

00:09:12: So you know, chat GPT and so on, all these kind of things.

00:09:15: So they are cool, but they are a little bit, I worry a little bit because people don't

00:09:20: realize that they are actually giving their data away when they use this kind of tools.

00:09:25: So they don't realize that actually the code that they asked the tool to build for them

00:09:29: includes some actually highly confidential information about their database or their

00:09:34: data or something like that.

00:09:35: So that kind of worries me a bit.

00:09:37: And secondly, it's not a genius.

00:09:40: It's just a tool that produces words, a text.

00:09:46: So it can just be giving you references to documentation or packages or whatever that

00:09:52: don't even exist.

00:09:53: Yeah, I heard about that too.

00:09:56: Yeah, so it happens very easily.

00:09:58: So it gets confused quite easily because it's not really that intelligent.

00:10:02: But we are going to that way all the time.

00:10:04: So there are now we have, we have a GPT-4, we have other things that are already available,

00:10:10: not just the three that I was mentioning earlier.

00:10:13: But I think we are very far with it, but not as far as people seem to think.

00:10:21: So it's not as intelligent as some people think it is.

00:10:24: But yes, I think machine learning is something that everybody should know about.

00:10:28: So if you don't know about it, you are losing a lot.

00:10:32: So yeah, we are talking about ML and nowadays, when we talk about JetGPT and so on, everyone

00:10:39: is talking about AI.

00:10:41: And I think everyone should know that one is not the same as the other, I think, but

00:10:47: the groundwork beyond that might be the same, right?

00:10:50: So we had a couple of episodes about the chat GPT stuff

00:10:56: and the GitHub co-pilot and everything

00:10:59: that is possible today for our vision, right?

00:11:02: So you're typing some text and you've got answers

00:11:04: and how impressive is this?

00:11:05: And this is the way why people are getting their hands

00:11:09: on machine learning parts,

00:11:11: just a little bit, the tip of the iceberg.

00:11:14: But I'm interested how you can use machine learning

00:11:18: in a corporate environment, not just for the end user who's

00:11:20: writing, texting, get some documentation summarized

00:11:23: or something.

00:11:25: And I think you could give us some insight how

00:11:28: you can use machine learning to, yes,

00:11:30: bring your corporate business logic to better results.

00:11:35: Yeah, so probably your data is in Oracle database, I assume.

00:11:39: Sure.

00:11:40: So then you have plenty of options.

00:11:42: So we have machine learning in the database.

00:11:45: We have there, we have, you're able to do it with SQL and PL/SQL.

00:11:49: You're able to do it with Python and you're able to do it with R. So you can create, you

00:11:54: can train, you can use any kind of models from the database, just like that.

00:12:00: But how you can do it, you can use PL/SQL packages, which is something that you probably

00:12:04: like because you are familiar with PL/SQL.

00:12:07: But if you're not, there are other options as well.

00:12:10: So you can use the thing called AutoML, which is a quite simple tool in the Oracle Cloud

00:12:15: Infrastructure.

00:12:16: So you just select the dataset that you're going to use and you define what's the target

00:12:22: that you want to predict.

00:12:23: Let's say you want to predict affinity card, for example, one of the classical examples.

00:12:28: So will this customer take an affinity card or not?

00:12:31: So you select that as a target, then you select this classification problem.

00:12:37: Then you define the ID, the identifier of the customer, and that's about it.

00:12:43: You press start the process, and it will create you the models.

00:12:46: So for each algorithm, it has in a database.

00:12:49: So for classification, it has plenty of algorithms.

00:12:52: For each algorithm and for your data set, it creates a model.

00:12:55: So you will have as many models as algorithms you chose.

00:12:59: And based on the metrics that you have chosen, so it gives you a list of possible metrics

00:13:03: that are for classification problem.

00:13:06: You choose one metric and based on this metric,

00:13:09: it will just put them in order.

00:13:10: So the first one is the best and so on and so on.

00:13:13: You can also add more metrics.

00:13:15: If this one doesn't tell you enough,

00:13:17: or maybe two is just the same,

00:13:19: you can add more metrics and compare those to each other

00:13:22: and decide which one is the one that you want to use.

00:13:24: And you just deploy it.

00:13:26: - Just click and just provide everything.

00:13:28: That sounds way too easy for me.

00:13:30: - It is, it is super easy.

00:13:32: And all these models that it creates

00:13:35: are stored in a database.

00:13:36: So from APEX perspective, it's very easy to call them.

00:13:39: You just use the apply procedure to call them.

00:13:42: - Okay.

00:13:43: - Very simple.

00:13:44: - So, and how do you train those models afterwards?

00:13:47: So you just say AutoML,

00:13:49: you told me the classification starts,

00:13:51: the models are being created,

00:13:54: and then it's working on the real data we have

00:13:56: and which will grow and grow and grow, right?

00:13:59: - Yes.

00:13:59: - And then this model will grow also.

00:14:01: - Yes, and Oracle also have now new tools

00:14:03: for actually following both the model and the data.

00:14:07: So if the model changes too much,

00:14:09: it will tell you and you need to train it more.

00:14:12: Or if the data changes so that the model

00:14:13: is not good for that data anymore,

00:14:15: it will also warn you about that.

00:14:17: So that's kind of a new thing.

00:14:18: Now it's only in a way that you have to call yourself,

00:14:22: but I'm sure it's gonna be like a service

00:14:24: that you just go and click like AutoML.

00:14:26: That's probably coming soon.

00:14:27: So that is very useful.

00:14:30: - That sounds promising.

00:14:31: Maybe we have a deep dive in the 23c database,

00:14:36: which just came out.

00:14:37: And we had a talk with Jared Wentzel,

00:14:39: which was the last episode, and he said,

00:14:41: there are many nice new features.

00:14:43: And if we are expecting some more regarding ML,

00:14:46: that should be in there, I think.

00:14:47: So you talked about different tools we have to use that.

00:14:52: Now the question in the Oracle environment,

00:14:55: is this for free in your database license?

00:14:57: - Yeah, the database is free.

00:14:58: - Yes, the one that is in database is for free.

00:15:01: And also auto-analyze for free.

00:15:04: But there's also services that are not for free,

00:15:06: like a data science service,

00:15:08: which is a Python environment in the OCI,

00:15:11: where you can just build your machine learning models

00:15:14: using your favorite Python libraries.

00:15:16: - Oh, great.

00:15:17: - So that's actually quite interesting.

00:15:19: And then you can publish those environments and your models

00:15:22: and whatever you want to publish from there,

00:15:24: and you can use them from other tools from Oracle.

00:15:27: Like if you publish conda environment,

00:15:30: you can import that Oracle database.

00:15:31: So you will have this conda environment in your database.

00:15:34: Or if you have a model that you will publish from there,

00:15:39: you can call it from other tools as well.

00:15:42: - So you can exchange the environments

00:15:44: with everything you achieved.

00:15:46: - Yes, yeah.

00:15:47: I think the coolest thing with Oracle environment

00:15:50: is that you can just integrate all your tools

00:15:52: easily to each other.

00:15:53: So you can use whatever is the best

00:15:55: for this particular use case, you can just use this

00:15:59: and it's not wasted.

00:16:00: You can use it from others as well.

00:16:03: And one of the payable tools is also Analytics Cloud.

00:16:06: Or Analytics Cloud, that can also do AutoML nowadays.

00:16:09: It can do all kind of machine learning things.

00:16:11: Actually, I will talk about that tomorrow.

00:16:14: So there's so many machine learning things

00:16:16: in Analytics Cloud, which is good

00:16:18: because that's the tool for the business people as well.

00:16:21: Kind of the people who understand the data,

00:16:23: not just technology.

00:16:24: Well, they are probably not too techies,

00:16:26: but they understand the data.

00:16:28: So this is a tool for them to be able

00:16:31: to use machine learning,

00:16:32: either the models that the data scientists

00:16:34: has created in a database,

00:16:35: so they can use those models with their data,

00:16:38: or they can use AutoML or some other functionalities

00:16:41: in OAC to create their own models.

00:16:44: - And just adjust what they need to do,

00:16:47: be changed for their reason.

00:16:49: Okay, so as always,

00:16:51: When I hear about new special technology,

00:16:54: I have the technology, I want to try it,

00:16:57: and then I'm searching for a use case, right?

00:17:00: So I think you need very big data sets

00:17:04: to use ML properly or--

00:17:06: - Not really, not really.

00:17:08: Actually, if you think about it,

00:17:09: more important is that your data set has good quality data

00:17:12: than that you have billions of rows of data.

00:17:15: So if you have a lot of data that is not good quality,

00:17:18: it's actually bad for machine learning

00:17:19: So it's learning different wrong things.

00:17:22: So how do I get information about the quality of my data?

00:17:26: So I know it from a different perspective.

00:17:28: If I have good data, you have this third normal form.

00:17:34: I don't know the exact translation of that.

00:17:36: But you have everything in place, a good data model for proper use in APEX application

00:17:40: or something.

00:17:41: Is it also good for machine learning automatically?

00:17:44: Or do you have a view on data modeling that you might have some ML in the future so you

00:17:51: just get some best practices in there?

00:17:53: Yeah.

00:17:54: One of the biggest things is nulls.

00:17:56: So if your application allows nulls, you might end up having a lot of nulls in your database.

00:18:01: So what's the point?

00:18:02: Because it doesn't tell anything.

00:18:04: So that's probably the biggest problem with this kind of things.

00:18:09: Another problem I often see is that you have a column that you have used for a purpose

00:18:13: And one day you are just too lazy to add a new column.

00:18:16: So you start using the same column for another purpose.

00:18:20: So the data is not consistent throughout their time.

00:18:26: So that kind of things can really affect badly.

00:18:29: Do we have another example for us where you personally used ML to reach a goal in a project?

00:18:39: Well actually there's several use cases where machine learning has been used.

00:18:43: It depends on your business where it's most useful.

00:18:46: But if you want to start with auto ML,

00:18:47: which is the easy part, you have two things to do.

00:18:50: You can do classification problems or regression.

00:18:53: Classification is that you put things into classes.

00:18:56: Like for example, if you're sending offers

00:18:58: to your customers, you can predict which of these

00:19:01: will end up with a deal and which will not end up

00:19:04: with a deal.

00:19:05: So you can put more effort to those that you know

00:19:08: that most likely will end up goodly and not badly.

00:19:12: That sounds so surreal to me that a machine

00:19:14: will give me information about how a customer reacts.

00:19:18: I think that my customers, that I can have a very good view

00:19:22: on how they will react on what I'm sending them.

00:19:24: How does a machine do that?

00:19:27: - Yes.

00:19:28: - And that's a magical moment.

00:19:29: I do not know if I can trust something like that

00:19:33: because I don't know how this works,

00:19:36: how this machine is predicting something

00:19:39: based on what I'm giving it.

00:19:42: So maybe you can help me with that.

00:19:44: - Yes, so that's a thing that people want to see

00:19:47: what why was this customer chosen as not potential?

00:19:51: - Okay.

00:19:52: - And you can see it from the data.

00:19:53: So why did the model make this decision?

00:19:57: So that's something the model tells you if you so want,

00:20:00: or why was this customer chosen as a potential customer?

00:20:04: But the truth is that model is never 100% sure.

00:20:08: And if you end up with a model that claims to be 100% sure,

00:20:12: you can be sure that it's not true.

00:20:13: There's something very fishy there.

00:20:15: - So what you're doing is that you're feeding information

00:20:19: into the system, which is telling,

00:20:21: I was sending this offer to the customer.

00:20:23: Those are the metrics of the customer, for example.

00:20:26: And if something, or if this offer

00:20:29: is not being successful to the customer,

00:20:32: I'm giving the information to the system

00:20:35: and the system itself decides for the next customer

00:20:38: if it's the same condition or how can I just imagine?

00:20:43: - The model is built using the features

00:20:45: like in a database, the columns.

00:20:46: So what do you know about a customer?

00:20:48: So based on this customer information,

00:20:50: the model creates ideas about what features are the ones

00:20:55: that are important for this decision.

00:20:58: So for example, if the customer is already our customer,

00:21:02: it's probably a feature that will have high impact

00:21:06: on the decision and so on.

00:21:07: So based on these features, it decides if this customer is potential or not.

00:21:12: So it's not this particular, it's not the name of the customer or anything that defines

00:21:17: if the deal will be closed or not.

00:21:18: It's the features.

00:21:20: So customers who are similar to this have been signing the contract and customers who

00:21:26: are very much similar to the other side are those who will not sign the contract.

00:21:31: So your job is to find these features that will tell enough about the customer for the

00:21:36: model to make the prediction. Most likely the first time you try it will not give you

00:21:41: good predictions because you don't have enough features about the customer. You have to add

00:21:45: more data. What kind of customers these were. So in your data set you have customers who

00:21:50: ended up doing the deal and customers who did not. And based on this it builds the model.

00:21:57: And if this data that you use for building the model is not good enough or it doesn't

00:22:02: have enough information, the model is not working.

00:22:05: Okay. So we have things like the size of a company,

00:22:10: the geolocation of the company, maybe.

00:22:13: Yeah, could be.

00:22:14: And as much columns I have,

00:22:17: the better the model will be in the predictions afterwards.

00:22:21: And also the columns that you have, you might need to transform them somehow.

00:22:25: So maybe a classical example is about predicting the house price,

00:22:30: which is, by the way, regression.

00:22:32: So we were talking about classification.

00:22:33: Another very typical use cases is regression.

00:22:36: So for regression, we are trying to predict numbers, like for example, the house price.

00:22:41: So how much would I get from my house if I sell it?

00:22:44: So one example of transformation is like, if I know my GP coordinates for the house,

00:22:50: so this is where it locates, it means nothing.

00:22:52: Okay.

00:22:53: What do you do with the coordinates?

00:22:54: You can't compare like your coordinates are higher than mine.

00:22:57: So what?

00:22:57: Sure.

00:22:58: It doesn't affect the price, but you can, you can transform the coordinates

00:23:02: to something more useful like location, like it's in the city center, in a very

00:23:06: popular area, or it's by the beach or by the lake or in the middle of the forest,

00:23:12: in the middle of nowhere, no services close by, you can transform it something

00:23:16: like that, and then it's useful for the model.

00:23:18: So the model can say that this is in a high end area, good area, price is higher.

00:23:25: This is in the middle of nowhere, the price is lower.

00:23:27: So that's the way our future looks like, right?

00:23:30: So if you have any like EmoScout, a company who is just rating the different

00:23:36: prices for locations you can move in.

00:23:41: And then if every single customer or every single company is using ML

00:23:46: and future to predict how good this location or how good this house will be

00:23:51: on the market and how good to sell it, that would be way easier than one guy

00:23:56: who's just estimating that, right?

00:23:57: Yes.

00:23:58: Yes.

00:23:58: I'm seeing that.

00:23:59: And also if you think about from your business side,

00:24:02: if all your competitors are doing this and you are not,

00:24:07: who's gonna win the competition?

00:24:08: - Of course, yeah.

00:24:09: - Yeah, because you still have the people

00:24:11: with a pen and paper and they are trying to do things.

00:24:13: And those are just pressing a button

00:24:15: and the machine is doing so many things 24/7.

00:24:18: - Yeah, I mean, but there have to be people

00:24:20: to understand that, right?

00:24:22: - Absolutely.

00:24:22: - You cannot let the machine do it and trust on it.

00:24:25: - The worst is if you let the machine do it and trust it,

00:24:27: Because then probably the data is completely wrong.

00:24:31: It's not transformed correctly for the model, for the transformation from data

00:24:38: algorithm to the model.

00:24:39: So it's transformed wrongly to the model.

00:24:42: The model is doing something completely different.

00:24:44: Classical example.

00:24:45: So there was a model that was predicting if the picture is of a dog or a wolf.

00:24:51: And it was really accurate.

00:24:53: People were like amazed.

00:24:54: How can it be so accurate?

00:24:56: But instead of really knowing if it's a dog or a wolf,

00:24:59: it actually knew it's a grass or it's snow.

00:25:02: So wolf were on the snow and dogs were on the grass.

00:25:06: So when there was a dog on the snow,

00:25:07: it was saying it's a wolf

00:25:08: and it didn't even look like a wolf.

00:25:11: So like, what are you doing?

00:25:13: So the machine learned wrong things.

00:25:15: You were thinking it learned right things

00:25:17: 'cause it was always correct until the day.

00:25:20: So you can never be sure what it actually learns

00:25:24: unless you check what it learns.

00:25:26: So why are you making your decisions?

00:25:29: - So you would go so far to say

00:25:31: this machine learning models

00:25:36: and the old technology around that

00:25:38: should just assist human people,

00:25:41: human beings to reach their goals.

00:25:45: It should just assist and not be the core of your company

00:25:49: or your goal, right?

00:25:50: - Yes, it should be assisting you

00:25:51: and you make the decisions based on what you get.

00:25:54: - That's great.

00:25:55: - If the decisions are crazy,

00:25:56: like one of my favorite examples is that

00:26:00: when I went to Los Angeles, it started raining.

00:26:02: I have been there once and it rained.

00:26:05: So the computer would say 100% sure

00:26:07: that when Heli lands to Los Angeles, it starts raining.

00:26:10: 'Cause it's not very common that it rains in Los Angeles.

00:26:12: They asked actually me to come back

00:26:14: because of my super skills.

00:26:16: So it started raining.

00:26:17: So the machine would say that if you want Los Angeles

00:26:20: to have rain, let Heli come here and it will rain.

00:26:23: So this kind of decisions would be nonsense

00:26:26: because the computer doesn't have common sense.

00:26:28: It only has the data and the algorithm

00:26:31: and it comes to a conclusion.

00:26:32: But if the data is,

00:26:34: there's only one row saying, "Helly, Los Angeles rain."

00:26:37: It's a fact.

00:26:38: - Yeah, right.

00:26:39: - But if you see outside, it's raining.

00:26:41: So I might have skills.

00:26:42: - Yeah, I mean, we are in Oslo right now,

00:26:45: and end of April and it's now snowing outside, right?

00:26:48: So maybe there's some conclusion on that too,

00:26:50: that Helly is moving.

00:26:51: - I have super skills.

00:26:53: traveling to San Francisco, it rains and you're traveling to Oslo and it snows.

00:26:58: Yeah, could be. Who knows? Maybe the computer knows better than we.

00:27:02: We will see. We'll see what the future brings us and how this evolves.

00:27:06: So another topic I want to talk about is one thing that is mentioned as a special

00:27:13: project for us in our preparation. It's about the tracking of COVID-19 exposures

00:27:18: in Finland. This sounds so amazing for me, what I read about that. Maybe you can

00:27:22: just talk about that and tell us what did you do for this COVID tracking?

00:27:27: Yeah, that was a fun project.

00:27:29: That really was.

00:27:31: So I get a message in Twitter.

00:27:34: So somebody I used to work with like 20 years ago,

00:27:37: he sent me a message that we are trying to track COVID cases in Finland

00:27:41: and we are using a blog and an Excel sheet,

00:27:44: but it doesn't seem to work anymore because COVID is growing.

00:27:47: And this is too many cases for for our technology.

00:27:51: So do you know anybody who could help us?

00:27:53: And we have no, we have zero budget.

00:27:55: I was like, Hmm, okay.

00:27:58: That was 11 in the evening or something like that.

00:27:59: I said, okay, let me sleep now.

00:28:01: And I will get back to you in the morning.

00:28:02: So in the morning I sent him a message.

00:28:05: I said, yes, we'll do it for you.

00:28:06: We will use the Oracle cloud, which is free.

00:28:09: And, uh, I will ask people who would volunteer to help with this.

00:28:12: So we will, we'll use Apex and we will build you a system.

00:28:15: And that's what we did.

00:28:16: So 10 days later we went to production.

00:28:19: So 10 days.

00:28:21: Yeah, that's a typical success story of Apex, right?

00:28:24: So we have two applications, actually.

00:28:26: We have the one for those volunteers who enter their data.

00:28:29: So they have their own credentials.

00:28:32: So everybody's using their own credentials when they log in and they enter their data.

00:28:36: And then we have another application that is open for everybody.

00:28:39: So anybody can go and see the COVID cases.

00:28:43: So where they are and so on and so on.

00:28:45: And for me, the best thing was, well, okay, after the 10 days, we've been improving it.

00:28:49: So it was not the, it was just the first release.

00:28:51: My part was the spatial thing because I love the multi-modal database.

00:28:57: I love the having graph and spatial and all this kind of things in one single database.

00:29:01: So I wanted to do the spatial.

00:29:03: And so I did all kinds of maps for us to be visualizing where are the hotspots in Finland for

00:29:11: COVID.

00:29:12: Now we don't track in Finland anymore because it's just like that.

00:29:16: But if you want to go to the website, you can go and you just select dates that are like a year ago

00:29:21: And you can see how it works, but it was a very fun project really

00:29:25: It's a second project I heard around this pandemic

00:29:29: Topic that that apex just just did a really good job. It did and OCI pre-tier as well and and actually

00:29:37: the local news

00:29:39: announced this this web page one evening and

00:29:43: Since the two hours from the news, it was 10 o'clock.

00:29:47: From 10 to midnight, we had 300,000 users in the web page.

00:29:51: Free tier web page.

00:29:53: Oh my God, I was sweating.

00:29:55: So me and one of my people who was also working on that,

00:29:58: we were just messaging with WatchUp to each other.

00:30:00: I'm dying, I'm dying.

00:30:01: What's going on now?

00:30:02: What's going on?

00:30:03: It worked.

00:30:04: Does it scale automatically even on a free tier, right?

00:30:06: Yeah, it works very well.

00:30:08: So we were like, "Whew."

00:30:09: Because you never know what happens.

00:30:11: because 300,000 is a lot in two hours.

00:30:14: - Yeah, sure.

00:30:15: - So that was a sweaty evening.

00:30:17: - Yeah, sure, of course.

00:30:19: But I mean, if you do something just out of nowhere,

00:30:23: like, yes, I can do so,

00:30:24: and you just use 10 days to build an application

00:30:28: with 300,000 users in the first two hours,

00:30:31: I can just imagine how proud you should be about that.

00:30:34: - Yeah, that was great.

00:30:36: And we learned so much also about APEX

00:30:38: and about many other things.

00:30:39: So it was a very educational project.

00:30:43: - And you were just mentioned by,

00:30:45: in the blogs of Oracle, right?

00:30:46: They use that for promotion, of course.

00:30:49: But yeah, it's sort of like tracking COVID-19 exposures

00:30:53: in Finland using Oracle Cloud

00:30:54: to develop an exposure tracking app in 10 days.

00:30:58: So this was like the best promotion for the cloud, right?

00:31:02: - Yes, yes, yes.

00:31:04: And I was so proud about my people.

00:31:06: So the volunteers who worked on this were from my company

00:31:09: and they did it on their own time.

00:31:11: - Oh yeah.

00:31:12: That's the motivation behind the Apex community, right?

00:31:16: How many people worked on that application then?

00:31:18: - We were actually, I think the maximum was four,

00:31:22: but most of us was two or three, most of the time.

00:31:25: - And there wasn't time for a big setup, right?

00:31:27: And get everything in place properly,

00:31:29: like a good project is just, okay, we need to do it.

00:31:32: We need to do it now.

00:31:33: And Apex is a perfect platform to get even good results.

00:31:36: - Luckily we knew Apex beforehand,

00:31:38: So we didn't have to start from the scratch.

00:31:40: - Yeah, sure.

00:31:41: But the good data model should be in place, right?

00:31:43: - Yes, it should, it should, it should.

00:31:45: - Exactly.

00:31:46: Yes, so let's talk about this conference right here.

00:31:51: Maybe you can tell us,

00:31:52: what are you talking about at the conference here?

00:31:55: - Yes, so first I have three presentations here.

00:31:58: The first one was about AutoML and AI services.

00:32:02: So AI services is also a payable thing,

00:32:05: but it doesn't really cost much.

00:32:07: It's really cheap to use.

00:32:09: There's different kinds of services for different use cases like document recognition or image

00:32:14: recognition or object detection, that kind of things.

00:32:18: So you can use the pre-trained models, just put your own picture there and it will do

00:32:22: the trick for you.

00:32:23: And if it's not good enough, you can always use the pre-trained model and train it further

00:32:28: with your own data.

00:32:30: And these AI services you can call from Apex, from data science service, for analytics cloud,

00:32:35: of the other tools you can just call these services, either the pre-trained or the ones

00:32:39: that you have further trained. I don't know what's the right word, so I use further trained

00:32:44: now.

00:32:45: Yeah, that's all right.

00:32:46: That describes what it is. So first I was talking about those and I had my younger son

00:32:52: with me here, so he was co-presenting with me. He's also working for me now. He's been

00:32:56: working for a year now. So he's been doing Apex, he's been doing OCI, all kinds of machine

00:33:02: learning things and especially a kind of Raspberry Pi, Coral, that kind of thing.

00:33:07: So I have a few questions about that, but maybe, maybe go ahead and I will ask the questions

00:33:12: how it should be, how it is to work for his own mother.

00:33:17: And then the other presentation we just finished with my son was about AutoML.

00:33:21: So AutoML as it is in OCI, but also as it is in Analytics Cloud.

00:33:28: And later today we talk about Analytics Cloud and what kind of machine learning capabilities

00:33:32: it actually has. So three presentations.

00:33:34: Three presentations, two of them with your son or three?

00:33:37: I think he will be on also on the last one.

00:33:39: Yeah, great. So there are so many questions I have. So how is it to be the boss of your

00:33:45: own son in the company?

00:33:48: The first question is how is to be boss for your own husband? Because my husband also

00:33:52: works for the company.

00:33:53: Oh, that sounds like a great combination in family life, right?

00:33:58: Yeah.

00:33:59: Is it hard for you to separate the job and the private time?

00:34:05: Sometimes it might be.

00:34:07: Because my husband is also a share owner of the company.

00:34:10: That's easier because we have common interests on the company

00:34:13: and all that kind of thing. So that helps a lot.

00:34:15: And besides, when we met we were working also together.

00:34:18: So we have always been working together.

00:34:20: So I think that's not an issue.

00:34:22: But with the son it's a little bit more difficult

00:34:24: because then you have so many hats to wear.

00:34:27: Sure, yeah. So when mom is putting the hat on, she's my boss and she talks the other

00:34:34: way around and not the same as my mom.

00:34:36: And I warned him when he joined the company. I said, I'm going to be really tough because

00:34:42: I demand a lot from everybody, but I'm afraid I will demand more from you since you're my

00:34:46: son. And I just have to tell you now before you start, so you know what to expect.

00:34:52: But it's a great start in the working life, right?

00:34:55: So he doesn't have to find a good job or find the right place.

00:34:59: I mean, he might have been into your work already and knew what would come in this company,

00:35:07: what you would do.

00:35:08: But is this his free will to come to this conference and talk?

00:35:13: Or are you pushing him?

00:35:14: No, he wanted to come.

00:35:16: He likes presenting too.

00:35:18: That's great.

00:35:19: was a question of mine because it's one thing that someone is presenting for his own likelihood

00:35:28: like I want to present right now, I want to try it or if it's hey go present, it will

00:35:33: be your thing, you're a good speaker, you should do that and it's okay I'm going to

00:35:38: this conference and talk and I want to make sure that he's...

00:35:42: He likes presenting, my elder son doesn't like, I don't think he would like to go and

00:35:47: but the younger one does.

00:35:48: And he was actually presenting with me last summer already.

00:35:51: We went to the EMEA leaders meeting

00:35:53: and he was presenting there about this COVID application.

00:35:56: - Ah, yeah.

00:35:58: Yeah, great.

00:35:58: So, I mean, he's learning from the best, right?

00:36:00: So, he was practicing his part of the talk with you

00:36:04: and I'm sure that this goes out very well.

00:36:09: So, he's going to present in multiple conferences

00:36:13: in future alone, I think.

00:36:14: - Yeah, I think so, yeah.

00:36:15: So the path is already made and he will be in this galaxy with us in the further thing.

00:36:24: I hope so.

00:36:25: Yeah, we'll see.

00:36:26: I'm very looking forward to see more of him in the upcoming future.

00:36:31: So I will go over to our categories.

00:36:37: Those are very often asked to our guests to have some insight of yourself.

00:36:44: So let me start with the category hypothetically.

00:36:48: So if you could undo one technological trend in the past, what would it be?

00:36:56: Undo a trend.

00:36:57: It could be everything, right?

00:36:58: It could be software, it could be hardware, it could be like cars, electric cars or something

00:37:04: like that.

00:37:06: I don't think I would undo anything.

00:37:08: Nothing to undo.

00:37:09: No, I think everything is good because if it's a mistake, you just have to fix it.

00:37:13: You don't have to undo it, just fix it.

00:37:16: Yeah, great.

00:37:17: Everyone is thinking about, "Oh, what should I say about that?"

00:37:22: But yeah, just don't undo anything.

00:37:25: Yeah, just fix it.

00:37:26: Just fix it and make it better.

00:37:27: Yeah, because one of my super skills is making fast decisions,

00:37:30: and that's the kind of same thing.

00:37:31: So I make decisions very quickly,

00:37:33: and if it's a wrong decision, I change it.

00:37:35: So I kind of--

00:37:36: You fix it before someone else just made the decision, right?

00:37:39: Yeah.

00:37:40: Okay, so if you don't want to make something unhappened or undo something,

00:37:48: what would you like to invent or create in the technology sector, which was not invented before?

00:37:55: I think what is missing from this field is probably human skills.

00:38:04: So I think that's something that we could add more.

00:38:07: So people are good with their job, they know the technology super well,

00:38:11: but they are not very good with other people.

00:38:13: So I think kind of getting more respect and understanding of other opinions

00:38:19: and that kind of things would be a good thing.

00:38:21: That's one part of many conferences and I for myself can talk for Apex Connect,

00:38:27: for example, where I'm working in the organization team.

00:38:29: We are trying to get as many soft skill presentations in as possible

00:38:34: because it's not always the most important thing to just have this

00:38:37: technological part of the conference and you learn software stuff or

00:38:41: programmatically stuff but you also learn how to deal with like hypersensical,

00:38:47: hypersensible people, how to work with other minded people maybe,

00:38:53: just work good work together and how to respect their opinions and so on.

00:38:57: That's a very great point. So additionally, do you have anything in

00:39:03: your mind for the future, what you want to deal with? Like, now you're in the ML universe,

00:39:10: universe, right? And you have so many topics we are talking about, like data modeling back

00:39:14: in the days, and then now ML. Is there any untouched topic for you, you would like to

00:39:21: learn more in future?

00:39:22: Hmm. Well, anything to do with data, that's what I love to do. So probably what I would

00:39:29: like to work on next is analytics cloud. So I have been working with it and I have my

00:39:33: certificates and everything, but I would like to learn further with that. So kind of understanding

00:39:39: data better and being able to teach people to understand the data better. So I think

00:39:45: that would be the next thing.

00:39:47: And there will be data forever, right?

00:39:49: There will be data forever.

00:39:51: It's getting more and more and everything is relying on data today. Yeah, great. Wow.

00:39:57: So I mean, additionally to that, I have another question from that, it builds up perfectly.

00:40:02: So what would you estimate?

00:40:05: What will your daily work look like in 10 years?

00:40:08: It's just training models or just revisiting or reviewing results from machine learning

00:40:15: algorithms and models?

00:40:17: I love the question because I don't even know tomorrow.

00:40:19: I mean, this is pretty good.

00:40:21: Yeah, my work is always a surprise because I work in so many areas.

00:40:25: So like now I have several projects.

00:40:27: I have an exercise migration.

00:40:29: I have a data model project.

00:40:31: I have machine learning projects, several machine learning projects.

00:40:36: I'm running the company.

00:40:38: I'm presenting.

00:40:39: I'm writing books.

00:40:42: I'm working on my PhD, so hopefully I will finish that one day.

00:40:46: So I had kind of many things that are, so what I'm doing most of the time is prioritize

00:40:51: things.

00:40:52: Yeah.

00:40:53: manage the next day. I mean we can summarize on the point that your daily

00:40:58: work in 10 years would look like the same as today but with a PhD right? Yes.

00:41:03: That's how it would be. Great. So the next category is in private. So are you

00:41:09: satisfied with your work-life balance? This might be something you already

00:41:13: told with your family structure that everyone is working in the same area

00:41:17: even in the same company right? Are you happy with the work-life balance? Can you

00:41:22: shut off, shut off from work. Like, now work is done for today. And I'm just

00:41:28: Yeah, I have a problem here because my work is my hobby. I kind of love this

00:41:34: work. You might guess the most most heard answer. Yeah, because in this

00:41:39: category, because everyone is saying that, yeah, I might refer to to Mike.

00:41:43: Mike Becker, who was like, I think the episode before the last one, he was

00:41:50: saying I don't have a work-life balance, I'm having a work-life blending.

00:41:55: It's not a balance, it's just seeing what is currently more important to me

00:42:01: and I'm seeing that how my life is basically in my work.

00:42:06: It's everything inside, it's everything blended, right?

00:42:10: So I was guessing that.

00:42:11: So if you are owning a successful company,

00:42:13: you don't have that much life between, so separately from work.

00:42:17: But if you see your work as part of your free time and life and hobby stuff,

00:42:22: that is, I think, the most sufficient thing.

00:42:26: So, another question from in private.

00:42:29: Would you show us your screen time on your iPhone without blushing?

00:42:33: So, don't say you don't have an iPhone because this answer was silly.

00:42:37: Okay. So, are you the type of person who is just constantly watching on the screen?

00:42:43: Well, I work a lot from my phone for some reason.

00:42:46: So a lot of things I do from my phone.

00:42:48: So I could say that 50% phone, 50% laptop or something like that.

00:42:54: But I, my, this phone is not peeping all the time as you can hear it.

00:42:59: There's no sound.

00:43:00: So it's been silent since many years.

00:43:04: It's not making any sounds.

00:43:05: This is completely quiet.

00:43:07: But my Apple Watch is giving me the sound.

00:43:09: It's constantly vibrating, right?

00:43:10: Yes, but not from emails.

00:43:12: So I have emails quiet.

00:43:14: So I don't get notifications for emails.

00:43:16: So if you want to reach me fast, you have to use WhatsApp signal.

00:43:20: Yeah, the messenger.

00:43:22: One of these or Twitter private message.

00:43:25: Those I get, but everything else is muted.

00:43:27: So I will read them when I have the time, but otherwise I don't.

00:43:32: Great.

00:43:33: So this leads me to the consumption part, the consumption category of my questions.

00:43:39: So yeah, I mean, you answered that already.

00:43:42: So how do you deal with the growing flood of information via various channels?

00:43:47: News and information are coming in frequently.

00:43:49: So you say that your phone is basically muted.

00:43:52: It's the same on my phone.

00:43:53: I don't even know how my ringtone is sound like.

00:43:56: I don't know.

00:43:58: So you said your messaging apps are coming to your Apple Watch and you see it quite instantly.

00:44:05: Do you have any news apps or some like breaking news and everything just to be ahead of time?

00:44:11: No.

00:44:12: news when I have the time. Usually in the mornings I start with news. I see Twitter,

00:44:17: so I have in my feed I have some news and then I might go to some newspaper pages to

00:44:21: read more about. Yeah, just to have this time to read and to update yourself and then the

00:44:26: rest of the day you will not be notified about that. That's quite healthy I think. So and

00:44:32: how do you consume news and new knowledge for example? So reading newspapers or news

00:44:39: pages but how do you consume new knowledge? Is it newsletters like Twitter, reading books

00:44:46: or magazines or? Everything. Everything. Everything. So I try to follow so many areas because I

00:44:53: don't have just one specific area that I do because I do so many things so there's a lot

00:44:59: to follow. So I either read the articles in the morning during lunch break, during coffee

00:45:03: during the breaks or I email them to myself and I read when I have time.

00:45:09: Yeah, this is a strategy I tried in the past too, but I was never reading.

00:45:15: Yes, me either.

00:45:16: And then at some point I was using the service, it was called Read It Later and then later on it was called Pocket.

00:45:24: and every browser and app you can share over the share sheet and say put it in

00:45:32: my pocket and my pocket grows and grows and grows and at some point I was just

00:45:38: reading okay you have like a thousand six hundred articles to read and I just

00:45:42: began with the oldest one I said okay I don't need that inside knowledge about

00:45:47: Oracle 11g right now because news is out there already in 23 so this is this is

00:45:54: something I tried and never worked to read something later.

00:45:58: So unless I'm not consuming it right now, it will pass.

00:46:01: And I will catch up with you as well.

00:46:02: - Yeah, that's true.

00:46:03: Yeah, it's usually better to read immediately.

00:46:05: But if it's a good article, then I email it to myself

00:46:08: and I put the title something that I can just search

00:46:11: by the title.

00:46:12: - And then you have to remember that this is something

00:46:14: you want to read and not just, okay, put it there.

00:46:17: There might be some time to read it.

00:46:18: Just, no, I will read that next week, for example.

00:46:21: Right? - Yeah.

00:46:22: - All right.

00:46:23: So the last question from the consumption category.

00:46:26: Do you turn off your phone, your Apple Watch off at night?

00:46:31: Are you always reachable?

00:46:32: I'm never reachable during the night because this is muted.

00:46:35: This phone is muted and I don't have my watch when I sleep.

00:46:40: Even as a CEO of a successful company, you're not reachable at all time, right?

00:46:44: No.

00:46:45: Great.

00:46:46: I need to sleep.

00:46:47: If I don't sleep, I will be angry and nobody wants to work with me.

00:46:50: So it's much better I sleep.

00:46:52: Okay, I would say do it like this in the future.

00:46:57: So I was really enjoying talking to you, Hilly.

00:46:59: Thank you for joining us at Devs on Tape.

00:47:02: And we will be happy for a second part maybe in the future.

00:47:06: - Absolutely. - We have so much topics

00:47:07: to talk about and it will not fit in this one recording.

00:47:11: So thank you very much and have a nice further conference.

00:47:14: - You too, thank you. - Thank you.

00:47:15: (upbeat music)

00:47:18: (upbeat music)

00:47:21: (upbeat music)

00:47:23: (upbeat music)

00:47:26: [MUSIC PLAYING]