SSE / Datopian weekly sync - May 05

VIEW RECORDING - 54 mins (No highlights)

@0:01 - Leonardo Farias

Hello, hello Jeff.

@0:05 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Hi there, how's it going?

@0:08 - Jeff Alexander

Doing great.

@0:09 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

How about you, man?

@0:11 - Jeff Alexander

Yeah, not too bad. Thanks very much. Hey, did you have a nice weekend as well?

@0:17 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Yeah, it was nice enough, I would say. As you can see, my head is shining. Yep, I guess that's good.

@0:30 - Jeff Alexander

Just to say, I think we'll be probably joined by a few other team members today. I'd mentioned a couple just about this AI tool that you were going to demonstrate and if you were interested.

So yeah, definitely keen to see it. Thanks for sharing the link as well. Looks really interesting.

@0:48 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Yeah, I mean, you're welcome. More than welcome for sure. other people are more than welcome as well to join this meeting, especially those who might have interest in this tool.

So, yeah, let's wait for them and then we can kick off since we don't have any other hot items on our plate at the moment, except one issue Ethan reported and we fixed it.

Oh, perfect. No, that's good.

@1:18 - Jeff Alexander

It would be good to chat through that. Obviously, Ethan from this week is joining the data quality team, so no doubt going to be on a few calls in the future, I can imagine, but less involved with the data sharing work.

But yeah, no, it's a, it would be good to chat through that as well. Just to let you know, the pen testing work, that is still progressing, so requisition has been raised internally for that.

There's just a few other materials some members of the team need, so we'll, I'll chase that up. Thank you.

Michael. Hi, Mark. Thank you for that. And hello, Mark.

@2:02 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Hello, Michael. Welcome. Thank you for joining.

@2:06 - Mark Fenton

Hello. Okay. I see one more person is trying to…

@2:12 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Oh, Rich is also joining us. Hello, Zoe. Hello, Rich. Hello. Hello.

@2:20 - Rich Baker

Hello once again and welcome.

@2:24 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Do we wait anybody else?

@2:26 - Jeff Alexander

I don't think we've got anyone else joining the call today. I think this is just about everyone. I would just say there's probably just a couple other minor things that we're wanting to talk about.

This is just a regular stand up with Datopian. Just invited a couple of others along to this session, just because you'd mentioned that AI tool, which sounded really interesting.

I think a few others were interested in seeing it as well. Yeah. Okay.

@2:56 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

I mean, if we don't, if we are not waiting anybody else… I guess we can slowly kick off, but before we kick off with the presentation itself, I would really like to introduce myself since I'm not sure if Mark and Zoe had the opportunity to collaborate with me.

I know that we had some email correspondence with Michael, but for the sake of the future collaboration, of course, and any potential support, I think it would be good to understand who are you dealing with.

Alright, so my name is Nikola, feel free to call me Nik, and I'm project manager here at Datopian, and I'm specifically assigned to your organization to represent Datopian, and if you have any concerns, issues with the data portal itself, or you have anything that comes on your mind that we should take in consideration, just feel free to reach out to me, and I will be more than happy to address these things towards our team and plan with our team members, our next steps.

So once again, Thank you for joining, and when it comes to the Queryless, or to be more specific, the tool we would like to present you today, we think this is a great opportunity to start this discussion, because I know that your team is actively working on more strategic roadmap, and I know that your team members were mentioning some chatbots that are potentially AI-powered in the past, and I think that we, at the moment, do have a product that we developed internally, that also can be a standalone product, but also integrated within your existing data portal, and today, Zhao, who was one of the main developers on this feature, is going to present us all potential use cases that can be covered with this tool.

We also understand that your portal is a specific one. So please be aware that whatever we are presenting today is just a showcase of visibilities and, of course, the functionalities that this tool is offering, but everything can be customized per specific needs.

So please just keep this in mind because nothing is limited as we are going to present. And, of course, feel free to interrupt us during the presentation at any time if you have any questions, so we can address these immediately.

And, I guess, with that said, João, maybe you can start if others don't have anything to guess that.

@5:42 - João Demenech

Hey, everyone. Yep. I'm going to get started. I'm going to share my screen. Okay, so for this demo, I'm going to be using this demo portal that we have for Porto.js, and we have…

enabled the Queryless integration here. That's why in the left-hand side, we can see this SAI button. And if I click on that, it shows up this chat.

And here we can start interacting with the AI. Just to note, this chat is always visible and keeps its state.

So you can navigate around and still have the assistance of the AI. And also, it's aware of what we're looking at.

So this is also being passed as context to the AI. So from this point, I'm just going to ask about a specific data set that I'm looking for.

So I'm going to say I'm looking for data sets about happiness. So I'm the AI behind the scenes is interacting with the portal API.

So… Well, so far we haven't seen any hallucinations in terms of making up data sets and so on, because it has this grounding on the API itself.

Yeah, as you can see here, it said that it found one data set, which is the World Happiness Data Set 2020, and it's linking to that data set, which lives in this same portal.

So this is a relative link. Finally, yeah, I also updated the context here, so it knows that we're looking at this page.

Here we could explore the data set, maybe preview the data, take a look at the columns, and so on.

And we can then open the chat to talk more about this data set. So here I'm going to start asking some questions about the data itself, and I'm going to start with something simple.

Also, what are the top 10? And this is something that we can easily verify by sorting the data by this column, the letter score.

Okay, so we are expecting Finland, Denmark, Switzerland, Iceland, Norway. Yeah, Finland, Denmark, Switzerland, Iceland, and so on. And it also provided us this interactive chart with the top 10 happiest countries.

We could then ask something more complex. Maybe create a chart of the average happiness perception or region. And we can do that because there's this regional indicator column.

And I'm expecting that it's going to do aggregation by this column. And, yeah, calculate the average. And there we go.

next column. See The Yeah, this, I cannot easily show you that this is correct, but I have verified before with a question like this one, and it looks correct.

And the reason why this approach makes it very unlikely that there are hallucinations is that all the model can do is query the API and create SQL queries.

So I could ask it, for example, can you explain how you calculated that? And then it's going to show exactly how it calculated the, yeah, it didn't show the SQL query.

Let's ask for that. But yeah, it's explaining here, I run a SQL query against the happiness dataset that grouped all countries by their regional indicator and calculated the average letter score for each region.

And then it's sorted for highest average first. So let's see. Yeah, here it is. Yeah, so it's selecting the region, the average of the letter score.

And let's see. Yeah, group by regional indicator and order by average score. Yeah, descending. So this looks absolutely right.

I'm to ask something else now, just maybe even more complex. So what factors contribute the most to the happiness perception across all countries?

And we can ask this because, yeah, there are some columns about factors that impact in the happiness perception, like social support, healthy life expectancy, and so on.

And there we go. So keep… Drivers of happiness. Social support is the biggest factor according to this data. So it generated this table for us.

And also a bar chart with each of the factors. And yeah, it also has this breakdown with some insights.

Yeah. So like money matters, but less than connection. We can also ask like, what other indicators or insights can we calculate?

And it can suggest things based on this data. Once again, because it's completely grounded by the metadata in this dataset, and it can only query this data.

cannot do anything else. So yeah, here it's suggesting a few other insights that we could explore, like happiness versus wealth.

So Original variation, uncertainty, and so on. So let's ask for one more of these, maybe happiness versus wealth, and then I'm going to show you a different feature.

Let's just see. Yeah, so happiness versus wealth. Sorry, you have… Yeah, I'm just going to jump in with a question there around another query that we could potentially ask.

@12:43 - Michael Glass

And I was thinking, you know, data quality cares. Data quality is a big concern to a lot of our data consumers.

And I'm wondering if there was a question you could ask around if there's any data quality indicators or gaps that this data set needs.

It might come back and say, no, the data set's perfect because it's demo, you want the data set to be perfect, but it'd be good to see if it does come back with anything.

Yeah, I can try that.

@13:10 - João Demenech

We haven't really tweaked the AI integration to do data quality analysis, but I assume you're expecting something like it's going to look for no's, for values that look like anomalies and so on, right?

Yeah, exactly that, yeah. Yeah, let's give it a try. I'm aware you've just said that, it's not tweaked for that, so it's not a tool that's for that, it's more for insights and, you know, talking about including the data set, but yeah, let's see what it does.

Yeah, let's see what it comes up with, but just to flag, since you asked this question, we could tweak it or even have another version of the AI just for data quality, and then we can tweak it, like, with proper instructions.

Yeah. Because in this case, it's expecting more stuff about, like, generating charts and so on. Yeah. But let's see, maybe it can surprise us.

Oh, I suspect it's going to suggest something, but, yeah, maybe not something relevant. Let's see. That's fine.

@14:25 - Michael Glass

And like you say, you know, that is a case of we could tweak it. And two, it does have that consideration around data quality, or we want to add a button that says, you know, let's go to a data quality view sort of thing.

@14:38 - João Demenech

Yes, exactly. Yeah. Oh, so it failed. Maybe it's rate limiting. Let me try again. Maybe we're going to have to wait a few seconds in this case.

But anyway. Okay, let's try again. Yeah, meanwhile, I'm gonna ask it something different here. So let's open a random data set like this one.

Just one last feature that I want to show you. Can you create a shareable report based on this data?

Which is also something that Queryless can do. Okay, yeah, so this time it worked. Do you see any data quality issues in this data?

And check results. So overall, this data set is quite clean, but there are a few notable patterns to be aware of.

So there are no missing values. All countries have complete data for core metrics. No duplicates. Mathematically consistent, possible ranges, and uncertainty quantified.

So yeah, these are, I think, nice things that they tried out. I guess it was running SQL queries behind the scenes to check this.

So potential issues. It has high unexplained variance in crisis countries. Yeah, but this, this might be more about… It's more insight of, to the data collection.

Yeah.

@16:35 - Michael Glass

Which again, it's, it's tweaked towards providing insight and, you know, insight into the data and all that, rather than specifically data quality.

But yeah, no, it's given a good stab at what's good, but also then, you know, explain that a little bit further as well around why it's kind of considered new things to be potential issues or whatnot as well.

So yeah, thank you. Nice. Yep. Yep.

@16:59 - João Demenech

Yep. Let's see here. So yeah, last feature I wanted to show you is just shareable reports. So yeah, when you're talking with Querylis and maybe you're exploring a certain subject around that data, you can ask it to generate shareable reports and it's going to provide you a link to a report that is hosted somewhere else.

And you can just copy this URL and share with whoever you want to share the report with. Yeah, that's it.

Just to flag, everything you're seeing here is customizable. So we can tweak the user experience on the widget. We can tweak the behavior of the AI.

And yeah, that's it. That's what I wanted to demo. Thank you, Jao.

@17:58 - Jeff Alexander

See some, some… some… some… See

@18:00 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

I just want to quickly add, maybe it's worth mentioning the connectors that it can be connected with external systems like, for example, Teams, Slack, WebEx, Gchat, or something like that.

So please be aware that this can be also tweaked in order to be accessible through outer systems for your personal usage, especially, for example, if we add additional skills for the AI, for example, as you mentioned, for the quality control of the data and similar.

So it can be used differently outside the data portal itself, which is exposed to end customers. Yeah. Thank you.

@18:50 - João Demenech

By the way, this reminds me of the smart meter data that is in BigQuery. And we can certainly wire that up too, just to flag.

And as Nikolaus was saying, we could even support different channels here. Cause this is one channel having this widget on the website, but if you want, you can even have something like a Telegram channel and yeah, it would work just the same.

Okay, thank you. And I think next is Leo. Leo. Leo.

@19:31 - Leonardo Farias

Yeah. I would ask related to the, if we could use data from different pages to generate a report, for example, two resources that we have on portal.

Yes. So it can actually do that.

@19:53 - João Demenech

Let me just see if, maybe let's try with these two. So global temperature, time series and global fossil. Fossil fuel emissions.

Can you correlate the global temperature time series dataset with global fossil fuel emissions? Yeah, so I actually have tried this out.

I think it was with the same datasets. So it's going to figure out which columns it can use to join the data.

And it's going to figure out Yeah, overlapping time period. And then it can run SQL queries that include both datasets here.

So, yeah, seeing that I found a very strong positive correlation between global temperature rise and fossil fuel emissions from 1850 to 2010.

And even generated a chart here with emissions and temperature. So it's capable of joining happen. I we'll to we'll this this

Different data sets and different resources as well. So that was going to be one of my questions as well.

@21:08 - Michael Glass

Great.

@21:09 - Jeff Alexander

I think that's something that we've talked about for a while as well, the potential for creating bespoke data sets, which I think some users have commented they would be interested in.

And so just to understand that, I know you've mentioned there's the kind of reporting functionality. Could you ask it, for instance, to create, you know, to bring data together from different data sets into one file that could then be exported?

Is that possible as well? Yeah.

@21:43 - João Demenech

So you're asking about a report, not like this one, but that you would download? Yeah.

@21:52 - Jeff Alexander

Like, could you ask it, for instance, to say, I want a data set that combines basically this different types of data together?.ắc a lot of set you Like, Howard So

And then is it possible for them to export that, saying like a CSV file, for instance?

@22:06 - João Demenech

Yeah, we haven't implemented that, but it's feasible. Just the same as with the reports here. Because in this case, we are generating a HTML file and uploading that to a bucket.

And we could do just the same with CSV, for example. I can even test it. So can you show me this data as CSV?

Of course, here it's going to just generate formatted data, but we could allow you to export this as a file and upload that to the bucket and so on.

Yeah, so in this case, it's just using year, temperature, and emissions. And it's going a long way. I think that's all the data that was overlapping.

Yeah, so even in this case, technically… Typically, I can export to CSV here, but we could allow you to actually download that data.

Yeah, just a small comment.

@23:08 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

basically it can, this Quereless definitely supports connectors, so it can be, for example, connected with, for example, Microsoft Excel or Google Sheets.

And basically it can create offline files as well, but also shareable CSV if needed. So visibility is there, but at this particular moment in demo, we didn't use connectors.

@23:40 - João Demenech

I've got a question just then on the sort of security side of this.

@23:44 - Michael Glass

And so it can combine with other data sets that can reach out to other data sets in the portal.

In this demo, have you got all these data sets like underneath the sort of open or public access? Is there any data sets behind this where you've got it restricted?

вторzer I didletkerning in this If the AI assistant, if you asked the AI assistant, can you pull from this dataset and you didn't have permission, would it restrict that?

Yeah.

@24:08 - João Demenech

So in this demo specifically, all the datasets are public and the AI is not querying the API using an API key.

So even if there were private datasets, it wouldn't be able to reach them just because it doesn't have the API key.

But for instances where you actually have private data, we can control whether an API key is being used and it can even use a user's specific API key so that it will respect exactly what the user has access to.

But what I mean is we're not going to have something like a shared API key. No, no, no. Yeah.

So yeah, we respect the CKN access control model.

@25:00 - Michael Glass

Brilliant, thank you. And then another question was just kind of, do you have any sort of energy industry data sets or examples that we could look at here?

Because I'm aware you've got a lot of global sort of fossil fuel emissions sort of stuff, but is there anything where it's using energy data?

Because we've obviously got a lot of open energy data out there that you guys can consume from.

@25:23 - João Demenech

Yeah, so I actually have, um, I was, I was testing with some of the SSCN data. Brilliant. Uh, let me show you.

Let me just find the, the portal here. So it's, uh, yeah. So I'm not sure if the data sets I imported are the most interesting to you, uh, but let's give it a try.

Yeah, so just for testing, I created these datasets. So we have GSP Technical Limits, Ordinary Active Network Management, Distribution Future Energy Scenarios datasets, and Defaults.

I tested with this one. Let me open that up. And yeah, is that an interesting dataset or is there any other dataset you would rather use?

Mark, you could probably talk a lot already. I think this one's a pretty interesting one.

@26:41 - Mark Fenton

It should have, as far as I'm aware, least primary level defense splits that are relatively interesting, because they should have both a rich timeline perspective and some geographic diversity, depending on if you can plot them as well.

@26:58 - João Demenech

So, do you have any… Any specific questions you would like to ask the AI?

@27:05 - Zoe Farrell

Can it do things like, because what this data set should do is sort of give you a scenario based on what we believe the technology uptake of a certain projection is.

Could it give you, could you ask a question such as, if the uptake was X rather than Y, what would it be?

Or is it not that sophisticated? I didn't really get the question, to be honest, sorry.

@27:36 - João Demenech

Let me just increase my volume.

@27:38 - Zoe Farrell

That's okay, I'm probably not articulating it in a very good way. Mark, do you know what I'm trying to say?

@27:46 - Mark Fenton

Maybe say, could you show the top five LCT projection primaries for air conditioning, but include, say, a 10-20% variance plus…

and see that projection. That gives you a bit of a scenario window that may not be within the data.

If that works in three, could hypothesize what it would look like if it doubled at 2025 or 2035, for example.

Yeah.

@28:17 - Zoe Farrell

Could you show the top five LCT projections for conditioning that include?

@28:23 - Mark Fenton

And include, and just say, and include a 20% plus or minus variance in the final figure in the visual.

@28:35 - João Demenech

Minus 10% variance? Yes. Sorry for not getting that.

@28:40 - Mark Fenton

Is that plus or plus or plus 10 or minus 10% variance? So essentially, given the 20% window on the projection.

I see.

@28:50 - João Demenech

Yeah, well, I'm going to ask that. Uh, hopefully I got that question, right? Could I quickly clarify, actually? So are we able to…

@29:00 - Jeff Alexander

Because this is obviously focused on data that we are creating, data that's already available on the portal. Would it be able to put the data in context at all by looking at the insights from third-party, by combining the data that we have with insights from third-party data that's available openly online?

Yeah, that's a good question.

@29:25 - João Demenech

If you point the LOM to it, I think it would work. But by default, it shouldn't do web search by itself.

@29:36 - Mark Fenton

Yeah, I assume it's since you open web searching and opening up your condition because you don't necessarily know what it's going to join to then, which will add a degree of uncertainty.

@29:45 - Michael Glass

Yeah, it invites the risk of much wider hallucination or misinformation, isn't it? I suppose you could train it to go to trusted sites, couldn't you?

@29:54 - Rich Baker

Like, you know, code body, regulation, you know, a trusted source. That's a good point.

@30:01 - João Demenech

Yes, it could be done. We could have something like an allow list and yeah. Okay, so it responded with, I just, I'm not familiar with the data set.

Yeah, I need you to. Yeah, I mean, it seems to have given an envelope of that plus, minus 10% window.

@30:27 - Mark Fenton

I would have expected it to produce it on the timeline, but then we didn't explicitly say to give that as a lying part chart showing the years with the envelope, but it's broadly done.

I'll be asked it.

@30:44 - Michael Glass

Yeah, but yeah, like you say, be more, we would be, could be more specific in the query there. It's like engaging with any LLM that you need to be specific enough to get what you want.

It's done what you, we've asked it. It's the machine.

@31:01 - João Demenech

Yeah, and you could also provide more pre-built context to the model. So, for example, there could be something like, because in this case, we don't have all the metadata available, but if it had all the metadata, maybe it would provide better results.

Maybe it could even have something like a field specifically for OOMs, so that they have this extra context about how to use the data.

So, yeah, there are lots of possibilities in terms of having more context.

@31:41 - Michael Glass

Yeah, just kind of on that point, just before Jeff, you come in there. It's one of the projects we do have running where it's an innovation project where they're looking at things called AI orchestrators.

So, essentially, instructions for how an LLM should utilize. That data set which would probably provide more context so then you could start to have that attached to the data set which would provide the LLM with enough context so that when you do query it, it goes, here's everything that you need to know, now query me sort of thing.

@32:19 - Jeff Alexander

apologies, just wanted to ask if there were examples of this being utilized like currently, if this is being used by any other organizations, just if you could share those examples just so we can see how So it'd be interesting even just to see how it's displayed, if it's displayed any differently across these different organizations and think about how we might implement it ourselves.

Yeah, so right now we don't have examples of other organizations.

@32:50 - João Demenech

This is a new product of ours. We've been talking to some of our currently existing clients, but yeah. We haven't yet deployed it to production.

Yeah. Well, we have… It was… Sorry?

@33:08 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

I just wanted to mention it was launched a couple of weeks ago when I were basically a week before I mentioned to you, Jeff.

So it's basically a fresh product, but really high demanding because we're constantly trying to improve it even more with the learnings we're getting and the feedbacks we're getting from the clients.

@33:31 - Jeff Alexander

Okay. Now that's definitely interesting as well. Just a very quick follow-on from the apologies. know Rich's got to stand up as well because I know that looking at the other network companies like the ones that are using OpenDataSoft currently, I can see that they've got some AI insights now available from their portals just to check if you're familiar with them yourselves or if you understand what…

the differences are between the solution and what's been offered by some of these other organizations? Yeah.

@34:07 - João Demenech

I think I have seen the one for Open Data Soft. Let me see if I can find it. Yeah.

But, well, I'm not sure if I'm going to find the portal I was using, but if I remember correctly, it has like a button in the dataset page that you can…

It opens up like a model and then you can ask for an insight or something like that. Personally, my issue with that implementation was that I would rather have this chat history than having like this area where the visitor…

is generated and then it's replaced in the next, next time you ask something else and so on. Um, but yeah, I think we, we don't effectively have, um, a more deep, uh, comparison.

Uh, we have, we discuss most, mostly, uh, user experience and we, we, we think the user experience that we are providing here with, uh, our implementation is more interesting.

Yeah. But yeah, I w going to try to find the, let me say, Stu, and then we can compare that life.

And Maurice, thank you.

@35:37 - Jeff Alexander

Cool.

@35:38 - Rich Baker

And then Joe, just in terms of a good segue from the, from the chat history. Question I've got is in the background are all the queries or, you know, sort of prompts, are they stored somewhere so that we could then in theory, look at all the users, all the queries that have been asked and then sort of categorize them with what's like a top five most requested.

Data sets or query. Yeah.

@36:03 - João Demenech

Yes, that's one of the selling points is that you can actually analyze what's being asked because that might be interesting for budgeting, budget justification, sorry, for explaining how people are using your portal and so on because it's much more meaningful than when people download the data and you don't know what they did with that, right?

So you can actually figure out exactly what they are asking. And yeah, right now it's, I wish I could demo that to you, but it's kind of technical, but we want you all to make this data available to you in a very friendly way.

And yeah, but short answer is yes, we're keeping that history. Okay, great. Okay. Yeah, that's one thing we could potentially look at then is if you were utilizing that.

@37:00 - Rich Baker

If you saw that 90% of every user was asking for the same two data sets and sticking it together to get some insight, actually, why shouldn't we be providing that as a case?

You know, why are we allowing our users to have to create that themselves? We should probably look at that as a shopping list that we should do that.

@37:17 - Michael Glass

Yeah, it's very much keeping the user on the portal so we can track that usage and get more of that insight as well.

As you see, it says it's We have no idea. Right, all this sounds really exciting. I'm going to ask some of the harder questions.

Who pays for usage then? You know, because obviously there's a token allocation, there's limits, etc. How does that work?

@37:41 - João Demenech

Yeah, well, we could support bring your own key and usage would be on you, but we have subscription models that are priced per message.

They're not really priced per token with, and yeah. I'm not sure if Nikola shared that with you already, or if we should bring this up now, Nikola?

We can bring this up now.

@38:08 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Yeah.

@38:13 - João Demenech

So maybe you can share your screen. Nikola, don't know if you have that there. The top, yeah. Yeah, but we have three different plans with different usage limits.

And we also have the option of buying extra usage, if you really want to keep that going, if you're having more usage this month and so on.

Yeah, but let's…

@38:54 - Michael Glass

Yeah, because you've got a kind of egg test thing, obviously, and we're now getting the contractual financial details around our existing contracts.

So we do We need to get into that in this call. We've obviously got our existing support and maintenance contract.

So would it be a plan bundled in with that? And then if we need to pay for additional things, it'd be as and when.

@39:13 - João Demenech

Yeah, I can see it nodding, Nikola.

@39:19 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

I'm trying to bring this up, but I can't find it immediately. So I do apologize for that when it comes to the price.

But as, as, as, as Jo said, basically it is a subscription based. So it, uh, we can discuss how this can be managed from the financial contract perspective.

It can definitely be part of the maintenance support, uh, contracts we already, uh, have, uh, since other professional services are like one time, uh, charging, right?

And this is something that we can figure out depending, as Jo said, on the, on the way, how we want to implement it.

to use like your tokens when it comes to LLMs or you would like to proceed with our already customized plans.

So maybe after this meeting, I can send you over this plan and we can proceed from there. And what would I also like to suggest is from your side, if you already have outlined the use cases that you potentially see as candidates for, implementation as part of the implementation of this solution, that would be really great.

Because immediately we can actually check the feasibility and tell you completely transparently, okay, this is feasible and this is maybe not.

But so far from what I heard from this discussion, I didn't see any constraints or issues with implementing any or anything of what you already mentioned.

but just for the sake of being aware offorce The full scope or your expectations, it would be really nice to understand, but everything else, aside of what we were mentioning, you would like to achieve with this tool, because I think, especially since it's not only customable, but also we can learn it and teach it to do some other things outside of the current scope, it's going to be really, really nice for your portal.

Brilliant. That sounds really good. Thank you.

@41:31 - Jeff Alexander

And we've got, as Michael has mentioned there as well, output from a recent project that was looking into developing a similar kind of capability, so there's some use cases off the back of that, so we'll tie in with the people from that team and see if we can share any of those use cases to support this as well.

@41:53 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Yep. That would definitely be feasible and great. Thank you.

@42:02 - Michael Glass

One of the other questions I've got, and again, this would probably be then in plan agreements and all that sort of stuff is, and depending on how well the model is configured, if it does suggest anything sort of dangerous or unsafe, like because we've got asset underground locations and somebody might go, where can I dig?

Can they dig into the garden? And actually there's a big cable. Where does liability sit on that? Now, that may be more of a legal lawyer question, but I don't know if it's something you've got ready to hand.

@42:37 - João Demenech

Yeah, I'm not prepared for that question. I'll have to bring this up to our tech leader, and then I can come back to you.

Great, thank you. A completely valid question, to be honest.

@42:55 - Mark Fenton

On that one as well, I assume again, I assume you're going to have these, but obviously all your sna

And you're going to have a series of standardized guardrails, you're going to have content moderation, you know, ensure it's an input safe, all those considerations.

I assume that's something that we'll also be able to have input on because we'll probably have our own series of standardized guardrails just to ensure the output and the questions are appropriate based on the use case and the conditions.

And is that considered extra development work or is that just part of the ongoing piece? And that's totally fine.

@43:27 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

That will be definitely covered because that's the reason why we do have these subscription plans where we can configure everything according to your needs, especially when it comes to the guardrails.

And this is really, really important topic that I guess will require separate discussion when it comes to the scope of the guardrails we want to put in place.

But completely valid concern. Basically, I mean, if we put… In the whole system context that will cover everything, non-issues should be expected.

Of course, it's an AI. As we said, we don't expect it. It hallucinates a lot because we are limiting it.

But of course, if we put more than less, it's going to be safer soon. Yeah, and that was kind of one more question.

@44:23 - Mark Fenton

Disclaimers as well. Let's see. You don't have one at the moment, so I assume you'll have a boilerplate. These outputs are generated by AI.

Please validate against original source. There may be risk of hallucination, as you get with everything now, AI-related.

@44:40 - Michael Glass

No other questions from me. Anything else from yourself, Mark? No, I think between the group, most of ones I had have been asked, which is great.

@44:48 - Mark Fenton

So thank you.

ACTION ITEM: Email Jeff/Michael/Mark/Zoe: Queryless pricing, blog/videos, demo link, OpenDataSoft comparison; request use cases + refs - WATCH

@44:52 - João Demenech

Thank you. Thank you very much. So just for the end, I would really like to sum up

@45:00 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Our next steps. So, as we said, I will send the queryless pricing plans after this meeting, right after this meeting, so you can check it out.

And also, you can completely feel free to outline the potential use case you do see as a candidate for implementing this queryless solution.

And, of course, if you do have some other potential candidate for covering these use cases, and you do have concerns, okay, can we do that as well, please create a reference, and I will definitely make sure we review that ASAP and address these concerns backwards as soon as possible.

ACTION ITEM: Consult tech lead re: liability; email Michael/Jeff/Mark/Zoe summary + guardrail recs - WATCH

And, yeah, yeah, will definitely consult with our technical lead regarding the liability surroundings and suggesting stuff. Brian, can I ask one other quick questions?

@46:00 - Jeff Alexander

Personally, terms of, like, I know it's very difficult to say at the moment as well because we've not really got all of our requirements just yet, and it sounds like it can be quite a, it's like a flexible solution as well, but in terms of the length of time that you can, like, imagine it would take to implement something like this, do you have a, like, rough estimate or a rough idea of how long that might take or how much effort would be required?

Mm-hmm.

@46:26 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

So, Joe? Yeah.

@46:30 - João Demenech

It should, like, the setup and deployment shouldn't take that long. Maybe customization, extra customization and validation from your end and so on can take some more time because, of course, I assume you want to really test it very thoroughly before releasing to the public.

But I think, like, think, in two weeks, maybe. Maybe we could have this deployed for testing, and then from there on it's just customizing whatever is there to be customized.

@47:12 - Michael Glass

Yeah, I think the biggest dependency is probably more on our side of the fence here, around governance, around alignment with our sort of SSE group strategy around AI, data protection, privacy, and then those guardrails and sort of context for the model as well.

Getting all that drawn out is probably going be 90. Like Mark said in background to me, 90% of the effort is probably going to be all of that.

The deployment is 10%. Yeah, exactly.

@47:47 - João Demenech

By the way, I'm saying two weeks, even with a buffer, potentially even faster than that, but yeah.

@47:55 - Jeff Alexander

right. Bye. Okay.

@48:00 - Michael Glass

Yeah, because one of the things, I think, for us as well, Jeff, and just in context for you as well, the Datopian team is that I think this would probably be our first AI implementation externally.

So it'd be the first time you'd be offering that sort of AI tooling to our distribution customers. So again, it might go through an extra level of scrutiny if we were doing all the internal governance things on this as well, because it might get a lot of questions and checks and balances and everything.

So if we are going to deploy this, it's going to be a lot of effort on our side, which is not a bad thing.

All that's good stuff to happen to. Absolutely. It might take a little bit longer than we'd anticipate of turning it on tomorrow, sort of thing.

@48:47 - Jeff Alexander

Yeah, no, it'd be good to catch up then on what's required there and how long that might take as well and who needs to be involved.

Yeah, would be good to know that. And then if you can share as much as you can at the moment, a nickel on.

Like potential costs and any more details around the time required for testing, anything like that, that would be really useful as well.

We'll do.

@49:10 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

We'll do, definitely. But now it looks absolutely fantastic.

@49:16 - Jeff Alexander

Definitely something that we've discussed previously in terms of like our long-term roadmap for the portal, something that we've wanted to explore in the future.

So yeah, it would be great to see what we can do here. That sounds great.

@49:33 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

And of course, I will get back to you via email as we already agreed, but feel free to reach out to me if you have any other concerns or more specific questions that may ease the process a little bit or speed things a little bit when it comes to decision making, because of course, of course, we will be more than happy to address this.

you. We are really putting a lot of efforts in continuous improvements of this queryless system, and, of course, we are more than happy to customize it for anybody's needs, so as much details you provide us, it's going to be even easier for us to provide additional set of information that you may be interested in.

@50:32 - Michael Glass

Is there anything online in terms of, like, blog posts, press releases, that we can also have a read over and refer to as well?

@50:44 - João Demenech

We have a few blog posts and videos. I can share a list with you. That'd be great. Thank you.

Please do.

@50:57 - Michael Glass

Yep.

@51:00 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

That also in the email I'm going to send after the meeting, xiao, so we'll prepare for that. Thanks, Nik.

@51:12 - João Demenech

Yeah, and by the way, also feel free to test, to try out Quirlyss on the demo portal. Yeah, I can share this with you.

@51:25 - Michael Glass

Yeah, I think that'd be massively helpful when we're talking internally as well, just like kind of give it a bit of an example of that as well.

Mark, sorry, Mark didn't get a chance to do an introduction, but Mark is our AI and Innovation Manager within distribution, so he's all things AI, and is leading on that front for distribution internally, as well as Mark, I think, is starting to hold more of the hands externally as well, so.

@51:52 - Mark Fenton

Yeah. Yeah.

@51:58 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Once again, nice to meet you. That's for you all.

@52:03 - Mark Fenton

Thanks very much.

@52:04 - Michael Glass

Nothing else from me?

@52:08 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Okay, we have eight minutes left, so feel free to raise your voice if there's anything. No, I think that's everything from me as well.

@52:19 - Jeff Alexander

Thank you very much. Really appreciate it. Okay, don't be shy, as I will say so.

@52:25 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Thank you very much for your presence. I really do hope that you liked what you saw, and I will get back to you ASAP with relevant information.

Thank which we already discussed. In the meantime, as I already said, don't hesitate to reach out with new set of questions if there are any.

I will share in the email this demo portal as well, as well as some blog posts, so you can try this on your own and play around with the features.

Thank you. I will chat. you very Thank you. Thank Yeah, I guess we'll be in touch and meet once again the next week.

But if there is any kind of need for earlier meet, Jeff, just feel free to reach out and we will manage to set the new session again.

And yeah, I think that that's all. No other topics from our side as well. Thank you.

@53:24 - Zoe Farrell

Thanks very much. Thank you very much, Tim.

@53:27 - Jeff Alexander

I appreciate your presence. Bye-bye.

@53:28 - Nikola Mladenovic (Viderum Inc. (Trading as Datopian))

Bye. Thank you. Thank you, guys. Bye.