AI AND DATA ENGINEERING SPECIAL WORKSHOP MEETING, 31-07-2025
GEORGE STEVENSON …that’s involved because we actually want to get you to tell us what you’re doing as well. But we’re going to kick it off as in Active plan is going to kick it off explaining what we do because I think that’s the first part of it. And then we’ll ask you to contribute your things. We actually have got lots of relationships with other software companies. And the reason I’m mentioning this is that what we’re doing here is data engineering to connect things together. So, part of what we’re doing here is, this is an example where for a particular client, this is a fire stopping report that’s a PDF as a document. And in fact actually there’s 42 documents that go to make this up, each of which has got 150 pages.
So what we’ve been doing is looking at how we can take information from the locations in this PDF document. And that identifies the unique reference for that particular fire stopping asset. But it’s a real pain for asset managers to be able to work out which fire stopping assets are across that corridor. So, what we’ve done, we’ve used data engineering to actually go through and extract information for the locations from these. And we’ve then read that into the database, so there we’ve got the tag number for each item and its location. And Tim’s going to show you and example of how we’ve done that a bit later on.
But the key thing about it is that it means that all of those fire stopping assets can then be viewed in a single view. And as you’ll see in a moment we can also then overlay other information such as fire doors and things like that. So, this is just illustrating how by using data engineering and reading it into the database, the information that was locked in that PDF now becomes real, intelligent asset data. And this is very exciting for me. It’s a way of really unlocking information that’s otherwise just in a document that is difficult to manually interpret. So, what we can also do here is turn on the fire compartmentation drawings so you can actually see where that particular fire stopping asset is in the context of the rest of the plan. And also we can then look at the doors.
So, within that same spatial model we can then illustrate where the fire doors are and be able then to drill into the fire doors. And this has come from a door schedule and an ironmongery schedule that have now been glued together using data engineering, so that we’ve actually got that as a coherent set of asset information. And that means that from a building safety point of view we’ve actually got the safety critical assets and how they’re related to each other in a way that is very sort of joined up. it also means that we can hold documents against those items because we’ve got a product library again that holds that information.
I’m now going to hand over to Andrew Handley who can start showing the type of methodologies that he uses as a normal BIM user and probably somebody that would be similar to you.
ANDREW HANDLEY (shares screen). I’m going to start the trilogy of presentations with the normal user. So, what are normal users? Well, typically they would be the likes of architects, surveyors, project managers. The type of people within Active Plan, myself and Richard highlighted there, could equally be that Chris does get involved, those types of things as well. So, maybe also known as no code or low code user, so you get to hear from a lot of very clever people who use a lot of code to do very smart things. Me as a normal user, I can’t do that, that’s not my skill set. So solving basic problems, generally getting a job done to deliver efficiently, we are sorting or you might say engineering data.
Working with it in isolated discrete steps, normally in much larger workflows. And those deliverables can also be to experts for further data engineering or even AI development work. So, as a normal user I would say I’m currently only using AI as a subset as internet searching. I have no illusions about how basic my use of AI is. So, it’s data engineering by any other name because even as a normal user we are actually making the data accessible and useable for various data driven activities, including AI by individuals and organisations.
Probably the best and easiest example is the COBie schema. This has been around for a long time, it’s a way of reducing drawings, models, schedules down on into a very simple spreadsheet with interrelated worksheets. So what are the problems I’m going to look at? Well, construction is inherently unstructured and commonly everything is being created with a rather myopic view. I don’t generally tend to see others around this have got better at working together, but we do all tend to as project team members to have a very singular view of things. Teams are now routinely providing data, most commonly in COBie, but this data still has gaps and issues that need fixing.
So, the three examples that I’m going to look at in sort of increasing levels of apparent sophistication, are clients and users who want a bespoke presentation of their data. So, they’re perhaps unwilling or unable to work with COBie in its raw form. The problems of long project durations. So, building construction programmes tend to be measured in years, not in weeks or months. And the never ending recurring challenge of spatial coordination. So, what are some of the solutions to this? Well, to the first two it’s Excel and a bit of, beyond the basic one, a bit of VBA a bit of visual basic for applications. And the third one, the only real way to deal with this issue of coordination is actually directly in the model.
So, clients and users want a bespoke presentation of their data. Let’s say in this case we have a client who wants their asset codes, but they’re not willing to look in the attribute sheet, they want to be way back over in the types’ sheet. So all they simply want is these, taking back and displaying, so we can very simply just use simple functions and standard formulas to look up and send that back based upon the type names.
However, the client then later in the programme complicates it by saying I don’t think some of these codes are right, I want you to update them. So they then have to be fed back into the model and into the COBie data. Which means we then have to perhaps look at using another bit of standard, just a very simple if statement, to do a check of as they are doing those revisions is it matching what we’re seeing in the model? And then make sure that we’re feeding that back in so that the right information is in the right place.
So the next one along, project durations. This is using a bit of VBA macros. In this case it’s to do with the Uniclass classification tables which are updated quarterly. This means we’ve a change log change, change logs and revisions and these are updating four times a year, so we get quarterly updates. There are 15 tables, change logs plus revisions, that’s going to equal at least 120 data updates per annum that have to be dealt with. I don’t want to have to do that 120 times, I want to find a way of automating that and being efficient and getting the data accurate.
The nice thing is that the format of the tables has now been consistent for years, so a simple bit of VBA means we can run very, you can see they are very basic VBA scripts, deal with the change logs and deal with every time they issue updates to the actual tables. It’s not very impressive, it helps me. What it does mean is that the Excel is then up-to-date and can be applied in many different places. So that small change and automation results in an Excel worksheet…so you can feed that back into Revit, into Airs? 9 mins 46secs, COBie validators or our validator into PIM and even into AutoCAD. So, you could update your layer classifications, for example.
And the final one, the most visual. So, in Revit models are used for a diverse range of deliverables. But we get quite often a problem whereby the coordinates that have been put out for the COBie are not coordinated. I won’t go into the complexities and technicalities of why that happens. So, the data challenge here is that we want to make those changes without actually changing the underlying model. So, this visually is the problem, all of these lovely little markers, these coordinate points, they should all be in the same place. You can see that they’re not all in the same place. So, we need to find a way of fixing this or getting the data because we’re not trying to fix this, that’s not our job, it’s to get the data we need.
And this visually is the problem that we get. Here is the outline for all of the project and all of the spaces. Here are the points from one of the other files that don’t fit inside the architecture. So, we need to get all of these to appear to be here and get those values out so that we can actually use them without changing the model. Dynamo let us build a graph, convenient nodes. This is the low code, this is me as a non-coding person just plugging things together. So we start by going and getting all of the things that are linked in, gathering up all of the things that we need to get the coordinates for. We then build a copy of all of those things that we need to get the correct coordinates for to pass them through later.
And then we also go and get for both of the things, the thing that we’re in and the thing that we’re trying to correct, what are the base points? So we can do the maths of correcting that difference in the difference that we saw between the points. So what happens is we end up with these raw points when they were not correct. And when it’s ran we get all of the points within the boundaries, it’s now all up within the building. So when we export this out to Excel we end up with a collection of X, Y and Z values as in coordinates that are all correct and fit within the COBie correctly.
RICHARD FREER I want to ask a couple of questions. So, when you say normal user, is that normal user the person who is doing the data engineering? Or is it the normal user who’s going to be using the results of the data engineering?
ANDREW HANDLEY It’s the creator, in these examples it’s not the end user. So, I’m showing three examples of things that within my day to day job at Active Plan I find myself doing to provide people…increasingly, the first one is very visible to the client, they can articulate. The latter two, this example that’s still on the screen, that is not visible to the people who actually created the models, because I’m not the creator, I’m just fixing these underlying issues.
Because of the diverse range of jobs that the models and the output from models are serving on a project, it’s not realistic to go back and say to the creators you’ve got this wrong, because this is a very small part of a very big set of things that they’re doing that doesn’t work because of a very particular requirement. So, we create what is needed so we don’t have to go back and ask them to do things that would just consume too much time and still probably wouldn’t actually be, we’d spend more time trying to explain it than just coming up with solutions like this.
GEORGE STEVENSON Could I just ask the group how many people are using Dynamo? Put something in the chat.
ANDREW HANDLEY Or similar, we’re not singling out Reviteer or using GDL or any other comparator.
RICHARD FREER Pierre is using Dynamo, Brian is using none, Jonathon none. Mark is using Dynamo.
GEORGE STEVENSON It’s quite common, a lot of people who are BIM coordinators might use this. I think the important thing here is to recognise that the model may look perfect from a visual perspective. What we’re talking about here is the underlying data. And what Andrew is doing here is illustrating how he actually corrects the underlying data so that we’ve got asset data in the right location. So, it’s in the right spaces from an asset management point of view. Is that right, Andrew?
ANDREW HANDLEY Yeah, I’ll slightly contradict George and say that the problem here is that it’s a mixture of everything that George has just said, plus the limitations of things like the Revit COBie add-in and needing to come up with something that was practical and worked and not offering so many. Basically you cannot bey default define which origin points to use to create coordinates. It uses the commonly unseen internal origin that most of us modelling and delivering projects never get involved with and never look at and never think about. So, it is a very unique problem to COBie, but it’s not a reason not to use COBie.
GEORGE STEVENSON Is there anybody else that’s got any experience of this?
RICHARD FREER Does anybody believe that they’re actually using data engineering? Are people in their current companies currently doing what Andrew is doing? Or are they aware that this is going on?
PIERRE VENTER The part I’d go through Andrew, is that we need to stop doing things where we do it repeatedly and we don’t correct it at the root. And everybody then picks up on that never to have to do it again. So, the idea that when you take a Revit file and you create it, the internal origin is there. And there are certain things which are crucial that you have it in the same position because automatically things align, a few glues and do various manipulations.
My big concern is sending it back to the owner, the author and correcting it there because downstream it means that we’ve got certainty that it’s owned by somebody and that it’s in the correct position, contains the right data etc. So, it’s principally just getting that workflow right so that we don’t redo work because what’s happening is that you’ve done it there. When you pass it on to another party and they get the one which is the black version, suddenly they have to go through the same process and redevelop stuff. And to get rid of that, to make it owner and user, very clear.
GEORGE STEVENSON I think we agree with you, Pierre. But also pragmatically we’ve got to deliver projects, so…
MO SHAHA’A I think the one thing that I just wanted to say is there is probably something in between. So I'm seeing a very good number of companies, effectively like Excel exists in this world where it isn't a database but ends up resulting in all kinds of challenges. Revit is a database to a degree that then isn’t really optimising for these types of things. I’m seeing an in between where generally no code tools are starting to be used to remove a lot of effort that that takes. And then companies use everything that ranges from an air table and smartsheets to ? 18mins 46secs which is the product that we own.
But because these relationships are then baked in, that then removes the need to then model all of these type of things within Dynamo and the likes there of. I can talk through that at some point. I generally think where things are going is there’s going to be less data engineering, in this side of things at least, rather than more. And this has only been needed historically because people haven’t been using databases for things like COBie to begin with. So we’re stitching together after the fact, rather than getting connected data to begin with.
ANDREW HANDLEY To go back to the previous point, I think our view on this is from a lot of experience and projects now, it’s getting this embedded in the requirements. A lot of the issues that we’re showing here are because asset information is bolted on after the project’s started. So, being defensive about some of the people who are producing this, they are producing what they were asked to produce. The problems that we’re solving is because someone has invariably come along and said actually, I want asset information out of this.
And that wasn’t actually considered and built in to the requirements at the beginning. And therefore checks aren’t being done like are you all actually working to a setting out in a grid file for the whole project. Because the project is not viewed as important, the deliverables in a lot of cases are still within in the contract just give us a mountain of drawings. So a lot of consultants quite rightly turn around and say that isn’t a deliverable.
RICHARD FREER Let’s move on and we have a real, live data engineer with us, highly experienced, Chris Aldridge.
CHRIS ALDRIDGE I’m a data engineer and get involved in integration products for Active Plan. (shares screen). And I’m here to talk about Power Query which I like to think of as better than luck, but not quite magic. Power query sits between the random number generator that is Excel and the magic that AI promises to be and has the potential to be. So, it sits fairly comfortably in that middle ground. I’m still using it to fix the same sort of tools that Andrew’s working on, to fix the same data. So, an awful lot of this is about transforming data from one format into another format where it could be more useful or more readily consumed by other systems.
And Excel is great, I spent decades using it, but it’s still to easy to end up sharing broken data. Formulas can break, you can end up with a million columns. We’re all familiar with these REFs popping up when you delete things that you didn’t mean to. And it’s very hard when you have to be very disciplined to use it in a way that’s reusable as more things come and you end up starting from scratch. And if it’s a large file then you can almost forget about it, it grinds to a halt. You can save it before it’s finished processing the cells and that again leads to data corruption.
So, it’s great for small, fast jobs, but don’t let it anywhere near dates because as we all know that messes them up. This column here is a good example of that. These are actually two character codes that Excel has decided should be a date and that takes a while to sort out. So, that’s where Power Query comes in. Power Query is this data transformation tool that’s supplied by Microsoft. Now, if you’ve got Excel installed or Power BI it’s already installed, I’ve never come across an installation where it doesn’t get installed. So, it’s there and ready to go and I came across it thought my use of Power BI which is the presentation phase where you can show the benefits and show the value in the data that you’ve engineered.
It’s not a tool in the same way, it can look very excel like but it’s not a tool in the same way. It’s not about data entry, it’s not about going in and changing individual cells or doing lots of manual things. It’s about setting up repeatable processes that can deal with large files and you can get predictable outcome from. I’m now going to screen share live to take you through some of that. This is an illustration of what I was talking about before. I’ve imported this and there was two character codes down here that first of all Excel changed to full on dates. And I want to change the formatting to text, it changes it to numbers. So, that’s the kind of thing and you can lose hour trying to fix that.
We've got columns here like this one where we've got spaces at the beginning, and yes we can trim them. It's fairly easy to trim it. You can insert column. But then I’ve got this column that’s right now, delete this column. I’ve lost data. And here, we can quite easily split that over several columns. And if I missed that, I've suddenly lost data in my other columns. Although it’s easy to spot when you’re dealing with four lines, if you’re dealing with tens of thousands of lines they’re the sort of errors that you miss until you deliver it to the client and they’re asking why things aren’t lining up. Why you’re data is bad, effectively.
So, if I go through Power Query now and do a similar thing. I’ll access it via Excel, the difference is this time I’m going to come in by this data option and import it as a CSV file. So, you can see again, it’s jumped in and it’s tried to fix my codes by changing them to dates, which is the last thing in the world that I wanted it to do. But I can fix it this time. I can say instead of just loading it going to transform data and this is where it runs up the Power Query editor. Like I say, it’s still quite Excel like, but it allows me to do a lot of things a lot easier. So here, we’ve got this code that’s been converted to dates, but now I can go back in and just say no, actually that’s not a date, get rid of that.
So I’ve set that to text now and it’s got my original text instructions. So, particularly things like SFG20, that’s a nightmare if you try to process SFG20 related data in Excel. So, you can quickly get those back to the formats that you want them. If you want to trim a column like I did in the previous one, I can go up here and I can transform that and change into format, so I can trim it. And it will trim it in that column, I’ve not had to create another column and I’ve not had to create another formula or delete columns, it’s all there. And while I’m there, there’s a lot of this type of thing. Some of the things that would be a function and would be an extra column are now accessible straight from here, so I’ve capitalised that.
You can do things with, you can select columns like this and you can say add column…I’ll come back to the date one. But here is another example. This I want to break up in the same way that I broke the other one up. So, I can come in and I can do my delimiter option as before, colon. And rather than it overwriting the data it quite sensibly just inserts whatever columns that it needs in there. This is messed up, so I had a column in there that I didn’t spot and it split that time up. But what you’re doing by using Power Query is you’re building up a history of what you’ve done. And you can go back at any time and you can delete those steps or change what you want to do in them. So, in this instance I might go back and say I do want to do it on colon, but I only wanted it on the left most colon. And now I’ve got all of those codes in the first one and then I’ve got all of my strings in the second one.
Likewise, I’ve got version numbers on the ends of these strings, but some have got three words, some have got four words, some have got one word. But it’s very easy just to script that again but say do it on the right most delimiter. So I can add a set, click the right most one and split it on custom. And it’s pulled off the versions just from the end and left the other ones there. So, what I’m trying to show here is there’s a lot of things here that this does that work very well when you compare to what you’re doing in an Excel. And you can take that out further. So, this one here we’ve got columns with people’s names in it at the end. And I could split that because that’s not really very good for analysis.
But what you can do is instead of just splitting it, I can split it on each comma. And now I can say I want the results as rows. And what that would do is that will expand that out and give me a repeat of rows for each one of the elements in that individual comma separated list. And once you’ve got that all together, once you get your data there’s loads more that can do. These are just a few tastes so if you’ve used Excel extensively it should make you go oh, that’s going to save me hours. So here, I can save my output as main, but I can duplicate this. I could say this is my Chris data, and what I want to do here is filter this view just to things that are related to Chris. Then I might duplicate it again.
GEORGE STEVENSON Chris, if i just ask a couple of questions now, just to help people understand. The reason we’re doing this is that we manage for Sodexo all their master asset type libraries. They’ve got 3,500 standard asset types and they need to map each of those asset types to SFG20 and Uniclass and various other standards. And when they win a new contract, they’re generally given an asset list that has been created by the previous incumbent. And they have to make sense of that. The way in which that other data set is created is obviously not using their taxonomy and their classification.
So what Chris is doing regularly is helping them with that. Being able to take things that are in spreadsheets, in databases in some cases, and then gluing them together. So, what Chris is showing you here is how he goes about doing that. But obviously this is quite technical and we may be losing some of the people in the group. But the point of you being here is that you probably have got a problem statement, you’ve probably got an issue that you need help with that either what Andrew’s shown you, you may be doing yourselves. It may be that when you’ve done that you then want, for example, somebody like Chris to do something with something like Power Query.
The most important thing there is understanding what it is that you want to do. And that’s what we’d really like you as a group to be thinking about. You’ve got information, how would you like it glueing together? What type of outputs would you want? And then we can work together to bring that out. And we’re very happy to show you how to do it as well.
CHRIS ALDRIDGE Just to wrap this up. I thought I would be talking to more data engineering people, so that’s why I went granular. What you’ve done here is you’ve created this history and this history of the transformation that you’ve done on that file can be saved and applied to other files that are in the same format. And likewise I can go back and edit any of these steps, I can reorder them. It’s a repeatable process that isn’t cast in stone and at no point am I throwing data away. So if finish this off now and just say close and load, what I’ll get rather than that initial TSV file like I’ve got in the first example when I opened it, is that I just get those tables that I output.
So, there’s George’s tasks, there’s my tasks, there’s a separate table of all of the data that I was processing. So, rather than it coming in and it being ad hoc, there’s now a process involved and that process is repeatable. That’s me. That’s literally just the tip of the iceberg, Co-pilot is an excellent tool for getting help on how to do some of that stuff and Youtube has a whole raft of great tutorials that I could point you out if anyone is interested.
RICHARD FREER I’ll tell you what, because we use within BIM4Housing and BIM In Asset Management with all of our operations and recruitment and stuff a lot of Excel. And I just messaged Jiss will this help us? And he said yes, so you’re going to be getting a call.
PAUL WOMACK So, if you’ve got, apologies I’m going to call it a spreadsheet. But if you’ve got an asset so each line is an asset, will you be able to give that asset a unique reference number to assist with the Golden Thread?
CHRIS ALDRIDGE I don't think that's necessarily related to Power Query, but there are a number of ways that you could give a data set unique IDs at the beginning of each line if that's what you're after.
PAUL WOMACK Potentially yes, And then if there’s any changes that is then monitored and each change is recorded.
CHRIS ALDRIDGE Again, the monitoring and the changes is dependent on where that data set is being used, how it’s being used. But this would certainly play a role in making sure that the data was correct to start off with in whatever system you did decide to use.
GEORGE STEVENSON I think one of the things Paul, is that ideally the information should be created at the outset by the author in the right format following the right nomenclature and the right taxonomy. And that’s where people like Mo and Pierre would say this needs to be done properly at the outset and you shouldn’t need Excel. But the fact is that one, 95% of the projects we’re working with have already been built, so a lot of the information is already existing. So, you can’t go back to the author and say do it differently.
Also, to be frank, in many cases the expertise of the people that are actually creating the data isn’t adequately high from a data perspective because that isn’t what they’ve been trained to do, in many cases. So, to some extent what we’re doing here is we’re putting in place something that you can fix things after they’ve not been done properly.
RICHARD FREER So, it’s kind of a revision and correction?
GEORGE STEVENSON Yeah, that’s why it’s caused data engineering really. It's engineering, or reengineering the data.
RICHARD FREER Reengineering, that’s different though, isn’t it?
GEORGE STEVENSON Yeah, I guess so.
CHRIS ALDRIDGE Transforming is another word into that because I think that’s quite accurate. It's taking a set of data and making it into something else, whether that’s combining it with other data sets or cleaning it or fixing it. Anything like that, transformation is the word that I’d use.
RICHARD FREER Yeah, to me that makes more sense. Has anybody else got any comments on what Chris has been saying?
THEO OBAYORI I work at HB Reavis, I just want to reiterate what Chris said, i think Power Query is an excellent tool. I think one of the things to highlight and I don’t know whether you really drilled in on it Chris, but it remembers steps not functions. So because of that it’s extremely scaleable. If you go from a table which has three rows to a table that has a thousand rows, because it remembers the steps including the relationships between all of the tables that you’ve created, that’s still a single process.
And all you have to do is click refresh and it memorises everything that you’ve done before hand. And I find that incredibly useful on our projects because I’m the only one using it and I have to run a bunch of spreadsheets and update a bunch of spreadsheets. And it’s all one click for me because I’ve set it all up to refresh itself. Just to reiterate, it is a really powerful tool and it’s worth the learning.
PIERRE VENTER Chris, well done and the same thing with Theo. The problem is that we’re repeating the same problems throughout. So, basically what I’m trying to put forward here is that Mo’s database on Morta, or just a validation database, is to go and say what are the kinds of problem statements that we have, the problem case. And the methods that they should be done, exactly like Richard said earlier, is that they should be done at the beginning of the project, set up those definitions.
And it’s also to start investigating what are the problems that we’re getting with existing data, existing assets etc. to make them work? Those little steps, so that again Power Automate, Power Query, that whole set of power tools are incredibly powerful for doing certain things. But you don’t want to end up rebuilding it. The amount of wasted effort rebuilding it each and every time because of the little nuances really is a painful exercise.
GEORGE STEVENSON That was the point I’m making, Pierre. If we can actually identify what the problems are as a group, then we can collectively start to come up with solutions to that. And obviously the biggest solution is to make sure that people don’t make a mistake at the beginning. But in practice, when you’ve got a project where a BIM model, there’s no such thing as a BIM model, there’s probably 20 or 30 BIM models. And each of them has been created for a particular discipline, perhaps. And then you’ve got to federate all of that information and therefore it’s not as straightforward as that. You’re absolutely right that that’s what we should be doing though.
CHRIS ALDRIDGE It’s just to agree really. I’m completely behind getting the data right and getting the systems right, up front and in advance. The stuff that I’ve been through there is from more of the real world data arrives today and they want something back tomorrow. They’re not going to go back and retool their systems or processors, so it’s about getting something out. So, it’s purely a pragmatic way of working with data in the interim until…it’s a means to an end, I fully accept that.
RICHARD FREER We’re going to move on. Tim Aikin who has been developing artificial intelligence, working with artificial intelligence since before it got shortened to AI.
TIM AIKIN (shares screen). So, we’ve gone through some quite technical stuff to low code. What I’m going to do is give you two problems which use no code, but they give solutions. As with all processes we get a problem and in this case this was my problem. Somebody had said I’ve got 5 door schedules for a building, but one of the door schedules is missing. And that was the steel doors which, as you probably would appreciate if you’re in facility management, are quite important elements. But there were drawings for each one of the fire doors. We’ve got those as PDF files and that’s the fact, that’s just what we’ve got.
So, we couldn’t go back to the beginning, we couldn’t say could you produce it as a spreadsheet, we just had this. So, what was the process? And why did I want to transform it into a schedule? Well, in our system what we have is the ability to click on a door, get information about the door and connect to a schedule. So it makes absolute sense to have all of the doors as a schedule and then we can display the record. So that’s what we want to do, that’s the objective. So, how did I start? Well, I used my go to new buddy which in this case is ChatGPT, but I use a variety of their AI tools. Even ones that you can access, you can pay for them, they’ve got different models. And you can also use LLM studio which allows you to run it locally.
In this case what I did was took one of the PDFs and said right, is there anything I can read out of there? So, that was the query and then you can see what the results were. So, there was a bit of an opportunity here. So I thought right, OK, so we can read one drawing, could we get some more information out of it? So, what we want to be able to do is produce the spreadsheet. Now, it’s a series of conversations to get to a point where it does the job. And then you just say OK, now let’s put that into one prompt that will allow me to repeat that for anything else that I get again. So, that was the one that was used, that was the model 4o. The limitations of using the web interface is that in any one upload you can only upload 10 files.
So if you’ve got 20 or 30 you’re going to have to do it 2 or 3 times. But you can get all 30 files up and then process them all in one go. Or you can say keep adding it to a data table. And if you’re not getting the columns that you want it’s just about finessing it. It is a conversation, but at the end of the conversation you just get it to say right, give me all of that as one prompt, which is the great thing about using these things. So, no coding involved really. The second problem is one that I think just about everybody n facilities management has which is finding information out of whole series of records. And we just need a way to do it quickly.
And if we use project extranet and they’ve all got ways of filtering the data, but they can be quite complicated, it can be quite fiddly. And what we really want to do is just use an actual language query which is what we would do if we had an assistant. And we would just say just go and get me the fire drawings or fire strategy drawings, or find me any documents relating to fire doors. So, how do we do it? Well, in this case all I did was get to a point where I said well actually I want things to go and be listed out like this. So, I’m going to give you a spreadsheet and it’s going to be quite simple, but it’s going to have a lot of columns. But the only columns I’m really interested in are these. And I want to get it formatted like this.
Now, when you’re using something online like this you can’t actually access the files locally, so it can’t access. So, the link has got to be something that you can cut and paste. So, this was the output. They won’t allow you direct access for good reason. And again, it’s just a matter of refining the output. And then this is the initial prompt that I developed quite quickly and it was as simple as that. And those are the results, so it actually worked and you can just keep using it. So, that’s a go to thing that anybody can do. If you’ve just got a spreadsheet of record drawings that have got series of parameters. Maybe they’ve got the categories, the work packages the companies, maybe a bit of attribute data, title, a file name.
The language model can actually look through those and start to make connections, connections that you wouldn’t necessarily have thought of. And it isn’t going to be perfect, but it’s certainly going to get you through a record set of say 20,000 records and get you down to half a dozen which are probably relevant. Or in the case of what I was looking for ones that were just to do with the floor. And as I said, that’s the output. So, at the end of it this is the prompt that I got it to write as well. So once finishing the conversation, just get it to do the prompt. And that is me done as quickly as possible.
GEORGE STEVENSON Can I just explain that that’s the technique that Tim used to do what I showed right at the beginning. Which was being able to go through and interrogate PDF reports and then turn that into meaningful object data that is then held in the database that can then be used from an asset management perspective. So, as Chris was saying what we’re doing is we’re transforming information that’s been produced for a purpose.
TIM AIKIN Yeah, it’s a little bit more complicated than that, George. I don’t want to make people think that it can do absolutely anything. What it can do is it can guide you to be able to curate a solution, because that solution also ended up with developing code. And then also trying out different techniques as well. So, because I’m in research obviously we’re going to try out different methodologies and things like that and find out what works best. To give you an example, I was given a problem this morning and how somebody had started to create a solution. And within 5-10 minutes I had a new solution, which now they seem to have given me the problem.
It’s because what this new world of AI allows us to do is explore faster and more intelligently towards things and bring up things that we may never have known about, or not experienced. And if it does throw up things you don’t understand, you just say what does that mean? It's literally like that.
GEORGE STEVENSON The point I’m making Tim is that the techniques that you’ve shown, all three of you have shown, are things that we’re doing as Active Plan on a day to day basis to create better asset information. Now, if anybody on the call wants to make that transformation that’s something that we can help with. And I don’t mind pitching the fact that that’s something that we’re interested in doing. But importantly, we’re also interested in sharing how we’re doing it with this community so that we can work together. You can do some of it, we can do some of it. And the ultimate goal is having much more useable, findable asset information that can then be kept up-to-date.
RICHARD FREER One of the most valuable things that we’ve found in developing this community is that a cat can look at a King. You can have the owner of a development sitting next to the guy figuratively in a meeting, the guy who’s installing the fire doors. But you can also have two competitors figuratively sitting together, sharing information. So, we make no apology that sometimes we use specific companies to talk about things because it’s going to happen. If a company has got a solution we want to see it. And that applies to all companies. And I think George would say that, as would other companies that have been involved with BIM4Housing and BIM In Asset Management over the last 4 or 5 years, everybody benefits.
GEORGE STEVENSON Yeah, we’re working now with Mo from Morta and those are the sorts of things, that sort of collaboration that is absolutely critical.
RICHARD FREER Well, it’s changing the game, that’s the reality of the situation.
MAKS ULJANOVS I just wanted to mention on the previous example that was shown, there is a thing called Microsoft AI Hub, it’s been mentioned before in this chat. And you can use the form processor tool from them to effectively extract data from PDFs. You can basically kind of train it to identify specific labels and specific data fields and then you can then run Power Automate to connect basically, to extract it and to populate into Excel spreadsheets.
GEORGE STEVENSON If that’s something that you’re doing within Associated Architects then that would be really useful to get involved with as well.
MAKS ULJANOVS I did a mini demo where we’re taking an asbestos survey and they all come with all different forms and shapes and stuff. And we just kind of looked at extracting room number location, couple of bits, and it seemed to be using quite well. You can also process drawing numbers and all sorts of stuff. But it seemed to be relatively precise. It does however mix up Os and 0s as of letters and numbers, but overall it seems fairly reliable.
TIM AIKIN Yeah, I think what’s happening is what a lot of people don’t understand is what the AI is doing is it’s actually writing code underneath to be able to process stuff. Particularly if you’re giving it PDFs, the example I was given one where they’d put room schedules into a PDF as images. It was data, got turned into an image and then got pasted into a Word document that then got created into a PDF. But we can just reverse it back out now.
But what you’re doing there is using OCR. So, where you get the problem with zeros and Os is because of the OCR. But then, if you’re then processing it through LLMs, then it will recognise that I think that should be a zero because the rest of the format is like that. So, there’s good examples of that, where it actually just makes the logical jump which just standard programming processes don’t.
GEORGE STEVENSON If there’s anybody else like Maks that’s actually, within the practice, within their organisation they’ve got people that are doing this work, please put them in touch with us. Because we really want to get this community to start sharing experiences. Because the fact is that if everybody is solving the same problem using the same methods, that’s very inefficient. We can learn from each other if we can share those problems and then bring that back.
The second thing is that over the next two days we’ve got several more workshops coming up which are practical workshops that we’re interested in getting people involved in. We’ve got one at 3.30pm called Watney Market. And what Watney Market is, it’s one that we’re doing a scan to asset process over the next 2 or 3 months. Tomorrow morning we’ve got one on data centres where we’re doing the same type of exercise. And then we’ve got another one tomorrow afternoon on creating as-built models using 360 photography. And then we’ve also got one on the Golden Thread Taskforce which is very heavily subscribed, we’ve got a lot of people signed up for that.
MO SHANA’A I really, really liked what Tim put out and I was going to build on what Maks was saying, Because I think what’s interesting about things like ChatGPT is we can start exploring a potential use case. But then if you look at the use case that Tim was putting out, they’re highly repeatable. And I think that’s where AI agents are going to play a big role. So, every time a drawing is uploaded to CDE, for example, you’ll want to extract things like the title and other things. I just wanted to know if you’re exploring the agents.
TIM AIKIN Agents are going to absolutely transform things because when we’re looking at things like the fire & safety bill and the amount of processing that we have to do to actually put things together to show that you’re actually complying, it’s a real ball ache. And if you have an agent that was able to go through at granular levels, you wouldn’t say read these thousand PDF files. What you would do is say read this information that there’s different levels and then when you need to just interrogate the files to make sure the information is actually in there, or in another place. And start to create the network so that I can prove if I get audited, or even for my own company, just report to me how you’re progressing etc. So, that’s one example of how to use agents.
The other in the facilities management market is to use the agents to actually monitor how well the building is performing. We see all of this with the digital twins which are only good if you’ve got data coming back that has information about performance. And sometimes the performance is very, very granular. It’s like I’ve only got meter readings, that’s not going to tell you about occupancy, is it? You’re going to have to infer stuff unless you’re actually measuring, you’ve just got sensors that just measure foot flow through the entrance doors, for one example.
Or you’re able to measure things switchboards or at plug levels to see how often things are being used. Things like in housing where we’ve got huge issues of humidity and damp and mould, all because the tenants or the occupiers don’t know when to ventilate a space. But you’re not able to actually monitor it. There’s massive opportunities.
RICHARD FREER If you’re interested in having a 15-20 minute chat to talk about where you are and where you could be going with Andrew, Chris or Tim, put Andrew, Chris or Tim in the chat.
CHAT
Ricardo Filho (External)… What AI model are you using for the OCR steps ? Not sure if you can disclose that
Phull, Pav Singh (External)… Hi George, is there consistency in Data management where clients have come together to create a common approach? Or do clients' Employers information requirements vary?
Alex Bauer - Multivista (External)… Multivista are using an A.I system to record and provide using photographs an automated assessment showing location of all unsealed penetrations attached to the fire drawings
Ricardo Filho (External)… What do you think about changing from excel to MS Lists ?
Mo Shana'a (External)… I've seen companies try this but lists isn't optimised for these types of data sets & usually isn't good for 3rd party access which ends up being important
Paulo Rodrigues (External)… I am an architect, and in practice we often struggle to get users to value data and use it appropriately within models. Current design authoring software, such as Revit, is not fit for purpose when it comes to effective data management or providing a user-friendly UI/UX experience.
In reality, based on observations, only around 10% of users — such as architects — are comfortable using tools like Excel formulas, Dynamo, Power BI, and other low-code platforms such as Power Apps / Power Automate.
Ricardo Filho (External)… Paulo Rodrigues i am based in Ireland, i work for Savills real estate company on ESG data management Pierre Venter (External)v\\
Pierre Venter (External) ..Paul Womack, the requirement is the feed-back into the core model.
And should be done at the start of the project
Ricardo Filho (External).. parsing ...
i would recommend looking into AI HUB from power automate
Maks Uljanovs (External)… this is perfect task for MS AI form processor
Ricardo Filho (External).. Tim Aikin - do you train a model locally? is it not "dangerous" from a client confidentially perspective?
Maks Uljanovs (External),,, it is all under MS subscription - I did not read their T&Cs but based on the Copilot promise data not shared
Maks Uljanovs (External).. I have used the MS Power Automate AI form processor to extract asbestos data from pdf survey and right into MS list or Excel
Ravi Kiran (External).. We need a Knowledge graph of asset meta data to contextualise information for AI to respond accurately.
Ricardo Filho (External).. that would be a dream for Real Estate as well
Maks Uljanovs (External)..
Ricardo Filho (External)… i can show an example of this workflow ... i finished one for BERs in Ireland
Brian McKenna (External)… Hi Max can you share a link to the MS AI solution thanks
Pierre Venter (External),,, Maks Uljanovs, thanks. I use Bluebeam which has a capability to read forms, sheets to extract the info.
Brian McKenna (External),,, watney market really good got to go
Harry Jaff (External).. Yes please include me in all of those.
Maks Uljanovs (External),, If you have MS account, navigate to power Automate
