Watts in Your Data

Denis: Hi, and welcome to a new episode
of the Industrial Data Quality Podcast.

I'm your host, Denis Gontcharov, and
today I'm joined by John Walmsley.

Hi John.

Welcome to the show.

John: Denis.

Hi.

Great to be here.

Denis: Alright then, today John and
I will talk about a very relevant

topic, namely concrete applications
of AI in the heavy industry.

But before we delve in John, for those
that do not know you, can you tell us

a bit more about you and your work?

John: Certainly, certainly Denis.

So, my background, I
is, I come from physics.

First part of my career I spent in r
and d and production in semiconductors.

it was interesting, I, my experience
there was through the entire product

development cycle and so I would go
from building the very first module

and writing in my notebook the results
from that, that device putting it

into an Excel and then finally to
on-prem data and then in the cloud.

And I enjoyed the benefit as
it went along through that

process, so I didn't have to be.

Supervising it the whole way, but I
also encountered a lot of problems as

as people start entering data and trying
to gather useful insights from it.

From there I went to medical devices
did a lot of work with startups,

helping them launch new products,
developing the technology in it.

I think the coolest thing we did was
an autonomous capsule that traveled

through the GI and made a sample
and transmitted the results to a

base station and up to the cloud.

That was a lot of fun.

But one of the big things I learned from
that, as well as the repeated process

of how you develop something new that
is going to be used by somebody who did

not develop it, also how to handle risk.

The, the medical device world
is a place where you understand

everything you do has some sort of
risk and hazard associated with it.

And you need to be thinking that through
as you step into your, into your process.

And now I think as part of what
we're talking today is in my context

as CTO of Aluminate Technologies.

So Aluminate is an AI company
bringing data driven solutions

to a various established
industry facing new challenges.

And I, I guess we'll, we'll
talk about that more later.

But it's probably worth saying
that Aluminate was founded

and grown by Kite Company.

Creator Kite is a venture studio model.

Looks to large industry to find
problems that the industry is facing

that could be solved with a company
delivering a long-term solution that's

supported for that whole industry.

And out of that work is where
Aluminate came and we heard from

some of the large companies.

The challenges they were facing,
again, that we'll talk about and that

has brought Aluminate into being.

And I guess one other last thing to say
is, for me, this is all about clean tech.

I'm really interested in providing
solutions for large greenhouse

gas emitters to reduce their
footprint, while, of course improving

safety and their economic value.

So I'm looking forward
to the conversation.

Denis: Yeah, me too.

In fact, John and I go way back.

We met over a mutual friend
through the aluminum industry.

We met also on a conference in October
in Lyon a while back, and John and I

had plenty of discussions about data and
data quality in the aluminum industry.

We discussed a lot about unified
namespace and of course AI applications.

So I look very much forward to clarifying
what exactly we mean by AI and.

What I find very important for
this episode is that to make it

concrete, right, John, we get a
lot of AI messages bombarded from

nearly every application that we use.

So I was hoping we can really discuss
what AI actually does in the heavy

industry today and in the near future.

So maybe a good place to start
would be you could tell us something

about what Aluminate would be doing.

John: Yeah.

Yeah, we can, we can definitely do that.

And I think, talking about startups
and in, I think it, it just

amplifies your point in the world
of investments and startups, of

course, today everything is AI.

AI is, and everything,
and everything is ai.

But I very much appreciate you asking
the question because for those of us

who then have to deliver a solution
or procure a solution or use this,

it, it really helps if we have an
understanding of what we're going to get.

And so I, I can probably talk
a little bit about Aluminate.

Aluminate is is, I already
said an AI company.

we're working on delivering a
solution to the industry that takes.

Data, both existing data, and this is
key new data from new sensors from the

aluminum process to deliver insights
that are actionable for the people

who are responsible for that process.

And, and I can talk a little bit
later about how the industry is under

renewed pressure in a couple of areas
and how this how this helps with it.

But I think probably to, to break it down
a little bit, you can look at the, the

cycle that we will be traveling through.

So we have initial, augmented learning.

So machine learning if we think about
process that's delivering data it, today

what will happen, what typically happens
is the process is delivering data.

Everything's fine.

No one's really looking at it.

There's a problem.

Someone goes back and looks at it and
says, oh look, here's what happened.

I, I, I can see it clearly in the data.

So at its lowest level, something
that provides processing in a

data loop that gives an an insight
into, say there's anomaly here.

This is something that needs to be paid
attention to, has, has enormous value

because it's continuously watching.

And such things have existed in the past.

And for me maybe I'll come back to
the broader AI thing for me, but that

augmentation is, is a very valuable
place to start, particularly in a

process with a lot of data in inputs or
a lot of data outputs where the, things

don't go wrong very often, but when
they do go wrong, they go badly wrong.

so if there's a way of having some sort
of automated supervision, it really adds

a lot of value to the, to the process.

So our our view is if we have that.

And we can do that.

We can run off one sensor, we can do
all that augmented machine learning.

But really to start adding
value is more about having more

of the context of the process.

So it's multiple sensors where, first
of all, we have that that augmentation

as we bring multiple sensors together,
we have more opportunity to do some

interpreted intelligence around
the dynamics between those values.

In a way that a person might not
easily notice when they look at

it, it might take a lot of work, or
it's just beyond a person and, and

general statistical tools to notice.

And so now we have an, an engine that's
running on multiple streams and looking

at cross products and and, and generating
insights from the convolution of of

the signals, from those, those sensors.

of course, where we're really going.

And, where we we'd like to get to
is where we have some autonomous

optimization where based on these
inputs and these insights, rather than.

Having a person go and react
just a little bit too late, the

process itself reacts ahead of
time to to take the correct action.

And then, and this is a little
bit beyond, I think where we're at

today as a world is it is to be in
a position where we can have those,

have, have the process optimizing
itself to, on a, on an ongoing basis.

I, I think probably
there's a lot of trust.

to build before we get to that point.

But, but that is, that is where
it's going, and those to me

are the three levels of, of AI
when, when I think about it.

But, but of course.

thing is the topic, and you mentioned
Microsoft Word and everything else,

is these, a lot of work, a decades of
work has been done to generate large

language model predictive models and
there's an enormous amount of work

being done, an opportunity in simply
applying the proceeds of that work.

Into, into processes for to enable the
people who are doing, doing their job to

fill in forms more quickly or to to, to
summarize to summarize conditions, give

reports and and, and, and I would say,

a bunch of other areas that, that extend
beyond that, but are really about applying

the, the tools that exist that we use.

At this point, I think most people
have used some level of a, a large

language model every day to to improve
their to improve their productivity.

Denis: Yeah, a hundred percent.

There's two things for me that
personally really stood out.

In your discussion about AI
and especially that answer the

question, well, what is novel?

Because in the heavy industry, we've
been looking at data for a long time.

At trends, we have mean max values, but
you mentioned that now we are combining

multiple data streams across products.

So to me that's pretty novel that we are
now not only looking at one particular

value, we are making a convolution with.

All kinds of different sensor
values and other event data to

give you more rich information.

I, I guess that's what you meant
with augmented data, right?

John: Yeah, I call that one
that that's the interpreted

intelligence, that that level.

But yes, if you're, a number of different
insights or data streams together, I.

To look at the, look at the prediction
that you can gain from that.

That's, I, I think that's the next,
that's the, the next level up.

And.

For industries, and we can talk a
little bit about the aluminum industry.

The aluminum, aluminum industry is
not a heavily censored industry.

other industries have far met war
sensors, and in fact, it was interesting

to listen to recent podcast of yours
with Thomas Dhollander of Timeseer.

I really enjoyed.

He talked about he talked about
ambition of, of as being an important.

Thing to understand is you're tackling
a problem, but also the the time series

as a way of, of looking at the data.

But you need to have that data first
to look at, to look at the time series.

I guess I actually, I, on a side,
side note, I really enjoy the, the

power of looking at a time series.

you are it, it, I've, I've worked
with, with engineers in the past where

they'll tell you, oh, the CPK on this
process is, is they'll give us a number.

They say it's 2.6

or it's four, or it's some, some
value of the process capability

compared to, the, the specification.

I always like, if they're a young
engineer, I always like to ask

them, could you just give me a
histogram and show me it's a normal

distribution so that the statistics
that this is all resting on apply.

But, but in other times, if you simply
plot a time series of the data that you've

been gathering, you can see patterns that
it's a way of accessing your intuition.

And this is something.

As people, we have this intuition.

We, we as humans, we are heavily, we,
we, we learn very early and you can

ask our, I think it's probably, it's
pretty innate in us to be able to

notice patterns that change with time.

And so this is where the
augmentation comes, is.

We could notice things, things
happening, we don't, 'cause nothing's

happened for a very long time.

And this is again, a, a, a people
problem, as, as people we're, we're

trained, if nothing bad happened recently,
it's probably not going to happen now.

Which is an important filter to enable us
to be able to tackle the challenges of the

world and a production data environment.

But if, if, if we live in an environment
where we know something bad could happen

and it could be very expensive or, or
dangerous putting in place a reliable

observer is, is, is very high value.

And that's, and, and that's, that's
that great starting point and I think it

gives us a really nice, starting point.

A nice rounded point solid, a nice
solid baseline to work from as

we start talking to people in the
industry and how they would, they

would look to adopt these tools.

Denis: Mm-hmm.

Yeah, I think that's the second
point you mentioned about the

observer, and it can be the human,
the engineer, but nowadays you have

this whole trend of having agentic AI.

We have in a non-human agent thing
in the process, maybe even in the

loop, taking decisions on behalf of
the learnings in the data itself.

From this sense, maybe we can
focus on some concrete use cases.

It can be from the aluminum industry,
but also maybe from other sectors

which are perhaps more advanced when
it comes to digital technologies.

Have you ever seen an example
of an agent AI in the industry,

or if not that advanced?

Have you seen other examples of data
being used or AI being leveraged?

John: So, that's a great point.

The aluminum industry is pretty
light on this sort of stuff.

So far.

It's it, it's people based, people are,
there, is an actual agent doing it.

It's not an agent ai.

there are some beginnings of
work in using assistive devices.

So for instance if you look at
there's thermal imaging or visual,

even just, maybe just the simplest
case is visual imaging where using.

Existing AI models and tools to draw boxes
around potentially things that have been

have been labeled as potentially hazardous
or non-optimal events to draw the user's

attention to, to, to something happening.

And, and aluminum for instance,
it's a very, part of the aluminum

process and is, it's a thermal
balance is important in the process.

And so where the and maybe maybe
I'll just pause a minute and.

give a description of the
industry or the, the process.

So the process is the, the industry brings
in alumina dissolves it in a, in a very

high temperature and caustic solvent,
and then passes very, very high current

through that solution to electrolyze the
alumina into aluminum and then oxygen.

which reacts with, with one of
the anodes to to create, which

is carbon based, to create CO2.

And that's an interesting
challenge of the industry.

It would like to be producing a
lot less CO2 as a result of that.

If the process doesn't run great it can
also produce other greenhouse gases.

So that's, there's another.

Motivation for, for I improving it,

but the process itself instead, it,
it exists of large, kind of apartment

sized what are called pots filled
with this mixture with the large

currents, hundreds of thousands of amps
passing through each pot in series.

To turn alumina, which is a
ceramic if we, we all know what

most all know what alumina is,
like a coffee cup type material.

And reduce that into
that metal and and gas.

And so that's the, that's, that's
the process that's running.

And it, there's a lot of work has
been done over the years to develop

it and have it run very well in the
continuous state and in that continuous

state, thermal balance is important.

And so, but also equally access to
the, to the contents of the pot, to

the solution is also important for
supervision, for changing of the of

the odes delivering managing the.

Cover that's that's, that's in there.

As part of the access, there
are doors that sit over, over

the pot to keep the heat in.

'cause again, they're thermally balanced.

And so a simple thing that I,
is to monitor the state of those

doors if they're being removed
and not replaced properly, the

process is running suboptimally
and and it needs to be replaced.

And this is an example of
something that can be monitored

using these, these cameras along
a line production line, which.

Is can be up to a kilometer long.

so if you are if you're looking
to monitor and take action on

such a long production line.

Some, some shortcut to being able to
notice an anomalous event is is possible.

So that's a very trivial example of
something that's, that's happening, but

it's, you can see it's achievable and
it's working, and it, it's important

step to getting the industry used to the
idea that these, these tools have value.

They're not job threatening for
the people who are doing the work.

It just helps them do do a
better job and and it, and it

produces value all by itself.

I don't know.

You, you have no doubt
run into other examples.

Perhaps Denis.

Denis: yeah.

Even from the same industry, just for the
listener, because the process is quite

complex and quite niche, we will include a
link to a good video, I think, from Alcoa

that explains the process in a visual way.

But to answer your question, John, yes.

In on the presentation I saw in a
conference in Dubai, that was two years

ago in 2023 EGA, so that's Emirates Global
aluminum, or aluminum, depending in if

you're from Europe or Canada presented
one of their AI applications, which

was a camera mounted on such a crane
that operates and Hoovers above such a.

The production line, and this
camera allows to inspect the work

of operators and detect potential
problems before the otherwise.

Because as you mentioned in
the beginning of this podcast,

detecting an issue in production
before it's visible in afterwards.

Save a lot of money.

John: Absolutely.

And, and I think, they, they, it's
their company, so they're allowed

to say aluminum and uh, and note,
note, note to the audience as I

actually grew up in the uk so I
started saying aluminum and then I've

learned to say aluminum and I try and.

I, I, the industry itself is completely
agnostic that the debate has been

settled, an agreement to disagree
and it's, it's completely bilingual,

but outside the industry it's can
be an interesting conversation.

Denis: Yeah, for sure.

So, John, one thing I noticed, I mean
I've worked for several companies

also in the aluminum industry.

Mostly supporting the data infrastructure
that powers these use cases that

feeds the data into the AI models.

But what I keep seeing time on time
again, is that we seem to be stuck in

this endless loop of pilot projects.

Like we have a project that can
either work or it cannot work.

But at the end of the day, I have
not really seen AI being rolled

out across an entire enterprise.

I would say even the.

LLM's, like the large language
models are the first time I really

see AI making a very big splash.

Do you feel you have the same experience?

John: Absolutely, absolutely Denis.

And we can maybe talk a little bit about
what the challenges of digital, I think

you want digital initiatives in heavy
industry, but the reason for pilots is if

you run a pilot and demonstrate it works.

That sounds great.

Everyone is, every you've, you've
cleared the hurdle that yes, this.

Generates seems like it delivers it
delivers the benefit it promised,

and it doesn't cause a big problem.

the next question is, is there
an incentive to roll this out

across the entire facility
and then across the industry?

There's, there's a large
potential barrier to doing that.

Lar you know, major projects
can and major projects can fail.

And more typically, major projects fail.

And so if there, there's,
there's, there's a natural

nervousness about about doing that.

And so people will want
to move incrementally.

I would say, oftentimes what what
clears that potential barrier is

the value that's being delivered
and the obviousness of the value.

So a large language model implementation,
if you think about this, this is if,

if you're implementing solution that
delivers that in an industrial setting.

One of the barriers in industrial setting
to rollout is that people, there's a

perception that the that the shop floor
people are not technically interested.

Potentially not technically capable in
terms of, they're very, very strong in

their process, but in terms of computers
and stuff like that it seemed like we

need to wait for the next generation.

Of course, you then go and ask
those same people, well, what's

in your pocket when you go home?

It's, it's your phone.

These people, they've got a phone as well
as the usual life wasting activities.

They're communicating with, with
people clearly around the world.

They're tracking maybe their stocks.

They're they're looking at they're
looking at any number of, of pieces

of data that will inform their
life and help them make decisions.

But when they get to work,
they're not given these tools.

And part of the reason, as I
say, is that, is that perception

that they're not ready for them.

And similarly, a concern of, of failure.

But if you think about
the large language model.

We're already seeing it being
de-risked in everybody's everyday life.

And so from, for, for those
people, they're not threatened

for the plant management.

There's likely some pretty
easy wins within reach.

And then at the board level first of
all, they're taking full advantage of

this new productivity generating tool.

Even if in time they
may discover it, it, it.

It doesn't do quite as much as they hope.

But it's a, it's a good place to
start and there's enough low hanging

fruit as, as people like to say
that it can have some success.

I, I think that's why
we're seeing it there.

But otherwise there's
the risk, value balance.

So if I roll this out, first of
all, there's all sorts of challenges

with rolling out a major initiative.

If, if it's too ambitious, it can fail.

If it has importance to a narrow group,
but not to the broader group, it can fail.

If everyone has a different
agenda associated it, it can fail.

you're crossing multiple
organizations, can fail.

And then there are new challenges
that will come typically

with, with new solutions.

And it can and, and if you're
not careful as well, your, your

implementation could be ill-conceived.

It could be poorly communicated,
under planned, under executed.

positive thing for heavy industry
is that generally the industry is

very strong in making positive,
high value CapEx decisions.

They have a good process for doing it.

They don't do it whimsically, which maybe
is one of the reasons things get stuck at

pilot, they have a strong decision process
and they have strong project management

processes for implementing broad once
they've decided to, to implement.

But they, when they do that,
they have to ask themselves,

who's going to look at this data?

And, and I probably the, the, the two
words to, to remember every time you think

you've got a great solution that will
generate more data for someone is so what?

And so, so, so what will this do you know?

Oh, it's interesting.

Yeah.

I think that's great.

And maybe your sponsor on the
project thought it was amazing.

Is this, is, this is exactly what
I wanted to know about the process.

I can now sleep at night.

Well.

they're not the general manager, the
fact that they can sleep at night maybe

isn't the most important part if they're
keeping the general manager awake.

Well, yes, that's certainly a
problem worth, worth solving.

the, so what is is, is so important.

And so the, I would say the reason
for projects getting stuck at Pilot

is that they, the value that they
propose to deliver is not sufficient

to clear that potential barrier.

For, for everyone involved to
take the risk of interruption,

the risk of investment, and the
risk of failure to roll it out.

Denis: I agree.

And this strong project management
skills and experience can actually

be quite an obstacle if you think
about iterating fast and scaling

something small to something big.

John: There, there's like, because,
know, but then on the other

hand, like it's a mindset thing.

'cause you can.

You can do a quick pilot on this.

I'll, I'll just keep talking about the
aluminum industry, but everyone, every,

everybody has, has multiple pieces of
equipment in their process, if that's

what we're talking about supervising.

But a, a trial on this pot can
now be a trial on two pots and

could be a trial on three pots.

And and so there where indeed where
if you can nimbly install things and,

and, and move them forward there,
you're looking for an organization

which does not have such a strong.

CapEx investment barrier and is open
to, to incrementally growing and

organically growing and implementation.

Denis: Yeah.

So you, you mentioned a couple
of interesting obstacles.

To, let's say scaling pilots.

The first one was more organizational,
probably more cultural.

Some challenges I also encountered, but
was more of my job as a data engineer when

I was really focused on getting the data.

Were also of a very technical nature.

It's often quite feasible to just
build this quick application for AI.

On the local system, but then you,
when you think about rolling out

this system across the entire plant,
or even about across all of your

plants in, in a continent or even
the world, at that point, you really

need to start asking questions
about your digital infrastructure.

And as you probably aware, dates
mostly from decades ago, right?

John: Yeah.

In your last podcast, which I would
again encourage people to listen to

you and Thomas talked about a lot,
the breadth of challenges of simply

getting data to flow in a, in a coherent
and accurate way from the point of

collection to the point of analysis.

Because if you, Denis are, have
a, have a project at a, a facility

and you implement something and
someone says, that's awesome.

Even the general manager says, that's
awesome, and you say, okay, well bye.,

I've made it work.

It works.

All you have to do now is
maintain it or implement it or.

it's, the subsequent investment
in maintenance as well is,

I think a huge factor here.

there is, I, to my mind, it
comes back to value and how often

people are so what, how, how often
people are looking at the results.

If you've got a solution, say it
works across the whole facility.

But it's doesn't have a, it
doesn't have a good architecture.

It doesn't have a robust architecture
that can be maintained going forward.

And as we all, we all know about,
or certainly we, software talks

a lot about code rot and the fact
that on a un unmaintained code.

In some narrow niches,
it can live forever.

But once you start connecting it to
modern tools the evolution and the

speed of development means that it
will, it will stop working fairly

soon if it's not maintained and the
maintenance may not be too challenging.

And of course, there's another opportunity
for AI to come in and provide some of that

maintenance, it's it we're, we're some
ways away from, from that at this point.

This is something we see, right?

And, and this is maybe the single
pot, the single trial as well.

Coming in with an
engineering level solution.

I worked for, for a, I think I told you,
a product development firm to, in medical

devices we were engineering services.

People would hire us to do a
thing, we would deliver what we

promised, and they would be happy.

And then they would turn that into
a product that they would support

and deliver to to customers.

this is something, sometimes, large
industry doesn't appreciate is when you

bring in experts to create a solution and
they make something that works, that only

lives as long as someone's interested
in it and someone is supporting it.

And this is the kite model.

This is where Aluminate comes from,
is we're not we're not set up to build

a solution, deliver it, and walk off.

We are making a product
that will be delivered.

To the industry and maintained, we will
have, we have a R&D roadmap with new

products coming forward as we do that.

New capabilities.

And and then similarly providing
ongoing and long-term support for the

product so that if there are issues,
they've got someone to phone, someone

comes and fixes it, they don't have
to ignore it or regret the investment.

And and each year.

There will be, or more often there
will be an improved functionality.

That's, that, that comes as part
of that part of that partnership.

And that's, the making sure
the value is there, but also

ensuring that there's a model.

The business model and the architecture
in place to sustain the solution over

the long haul are another critical part
part of having a successful solution.

Denis: I think it's a very interesting
idea to bring in external knowledge.

'cause we have to face, having
worked in the heavy industry,

I think they're experts.

And in fact, I would expect they
would have the answer to the question.

So what, they know their problems,
but they may not necessarily have the

technical skills or products, technologies
themselves to implement a solution.

So I really like the idea of
having external companies and

going back to my question about.

The problem of data infrastructure.

You have now very, very big
players like Databricks that solves

exactly that for large companies.

I think that what I expect and
hope to see in the future is a lot

more symbiosis between, let's say
the older industry within these

newer startups and newer products.

Do you see it also
evolving in this direction?

John: yes, I, I agree.

So that the, your Databricks example is a
great example where someone coming in and

implementing a solution could build their
own database structure on premise, and I.

There it is.

What, what Databricks is doing is exactly
what I'm saying, is creating a product

that has, of course, has functionality
that you can, you just buy it.

You don't have to develop it,
you just buy it, you use it.

It's maintained.

It gets better over time, not worse.

And there is, there is someone
to call if it stops working or if

it's, if you want to make, you want
to solve a new problem with it.

I think that the scale that comes from
that and the long term value that comes

from that is much, much higher than if
you can even attract a database admin to

build your own system or knit together
the various databases that that you've

inherited from your OT providers.

It's, it's I, I think just, and,
and this, the pace of change.

The pace of change is such that if you
can work with people who are working,

who are specializing in that area of
expertise, they will be on the cutting

edge and able to deliver good solutions
or maybe just I, in my head, as I said,

on the cutting edge, I thought just
behind the cutting edge for delivery

of solutions, but they're also working
at the cutting edge to understand

what comes next rather than hiring
a data, a a keen database admin who.

Implements at the cutting edge doesn't
really understand that it doesn't

fully have all the functionality
yet, and then it gives up in failure.

Or someone who implements something
that's, that's just very, very

old and very, very brittle
and, and, and not a solution.

So bringing that, bringing that
expertise in is, is an important part.

But I think if you are, if you're
relying, if you're going to rely and,

and if there's value there, you're
going to rely on this solution.

It wants to be, delivered as a
product by by a company that's,

that's in this for the long haul
and to, to keep keep delivering that

solution and supporting it, I think
is, is one of the keys, as you say.

Denis: Yeah, I see a future where
heavy industrials will have.

Purchased some form of professional
data platform that hides away

most of the difficult details of
data management and governance.

Actual calculations and analytics so
that they can literally stitch blocks

together and build data products
themselves, but also be able to purchase.

Additional products that they
can integrate easily on that

new system by just bolting it
on top from innovative startups.

I think we are moving away from this
phase where you had engineers writing

a lot of custom code in a project
and then building a use case that

might, may work for a while, but then
with that company, when that person

leaves, the company gets abandoned.

I really think we are
starting to mature in.

How we are building lasting data products.

But I wouldn't say we are quite
yet at this point where they're

being rolled out enterprise wide.

John: I agreed.

Yeah.

I think you know this, this is
where society is, is gaining a

big benefit from the SaaS model.

The i, the realization that you don't
have to build this from bottom up.

A lot of the architecture is,
is available to you and is,

is reliable and can be used.

And, and your key thing is to understand
the so what as described by your customer.

To turn that into a solution,
bring the pieces together.

And so in our case, again, Aluminate,
we were, we're bringing multiple

sensors, some of them existing
outside the industry, others new

to the world, never existed before.

That will provide useful
data out of the process.

That frankly, a lot of that, the, the
data just either isn't available today or

it's collected on an intermittent basis
by a person walking around with the, the

variability that's involved with that.

, And just the general intermittent nature.

It's a big part of, of solving the
problem is to bring sensors that do that

continuously, but being able to rely
on existing data infrastructure So that

doesn't all have to get built today.

And then it will also be it, it's, as new
sensors come forward, we can connect them.

I think is a very
enabling solution for us.

And then it's enabling solution for,
for customers as well, because then they

know that they're not relying on, on our
genius data architect who, we have, but

what it they in their heads, what if, what
if our genius data architect disappears?

Where are they left with?

And so we have the whole thing underlying
and relying on on, on some fairly proven

technologies while we do the really
clever part of, of applying that to

the industry in a way that's, that's
valuable to them and trustworthy.

Denis: Mm-hmm.

Yeah, we mentioned some
interesting applications already.

Just to repeat, like the visual
analysis pattern recognition.

I think the key that will still
be very important in the future

will be the human in the loop,
like the actual domain knowledge.

Otherwise, we risk rediscovering
a law that some French chemists

discovered in the 19th century,
something we essentially already knew.

So I think that will always
remain very important.

So what, what's your view on integrating
more human knowledge into these systems?

John: Our focus at the moment is
integrating insights with the human

knowledge that exists , so we've
been told and we're aware of key

parameters that people look for that
are an indication that they're going

to have an adverse event, and so we.

Sometimes some of these things are
noticeable when you look at the, if you

happen to be looking at the data at the
right time, you happen to be looking at

the at the hardware at the right time.

And so we can, we can, having been trained
in those regard by the, by the experience,

we can bring that into the system.

So then we're, we're delivering
insights that are the sea

insights they would've had, but.

the time, not just by
happenstance or or intermittently.

We, as we go forward, then of course we
should be providing more those, those

insights on a routine and regular basis.

But no one's really being paid
to do that work at this moment.

These, they just, they, it's
an incidental part of it.

What they're really being paid to do
is to make the process better and to

reimagine how the process will work.

And so in addition to
providing that ongoing.

We, I think we then get into our second
stage of interpretation where we're

interpreting multiple sensors to give
a, with a digital twin, for instance, to

give a, which, which exists today, right?

There's a 'cause I, I think
probably there's a risk that we

skip over the technical knowledge
that sits within the industry.

There's very deep electrochemical.

Knowledge of what's happening within
the process and how to optimize it.

And so making sure that we're
integrating with that and giving

feedback to it is is, is critical.

The, what, what we'd like to get
to in the end is to give, to give

those people a lot more tools.

So rather than being you know that they're
working around a local minimum or local

maximum in the process optimization,
we have a fully characterized process

with with, with recommendations,
there's an opportunity to do a lot

more broader, what if exploration on
how the process could be configured?

If you think about, so the
aluminum process, it's, it's

over a hundred years old.

It's had a, a number of
revolutionary changes in how it's

how it runs, but not that many.

it, it runs very well, very continuously.

It's facing new challenges.

We know this is, we, it's facing new
challenges where there's, the, is is,

producing CO2 is, is is something that's,
that's not ideal and they should, and the

industry is working on to, to reduce that.

other greenhouse, greenhouse
gases that get emitted as well

is something the industry is very
aggressively pursuing and looking to

understand ways of resolving that.

And they're also facing new
challenges around power availability.

I, I talked about the amount
of electricity that runs each

of these plants the power.

As we electrify, as a world, there's
clear opportunity for aluminum smelters

to be part of the solution of managing
supply for periods when when power,

power demand exceeds power supply.

this is a, this is a real, it's,
it's a new economic pressure on them.

And we see them.

We see them responding.

To that.

So there, there's real incentives to,
to take what's known about the process

today and to make it fundamentally
better and, and potentially different.

And I think that's an area where
again, we're taking our outputs and

integrating it with the human knowledge.

Denis: I think those are great
points and they directly answer

the question about, so what?

Right?

I think a lot of heavy industry
manufacturers got burned by ai, by the

initial promise of discovering secrets in
their data, which of course they haven't.

But what you're mentioning here, first
of all, the increased legislation that

demands lower CO2 already puts more
demands on their data and AI capabilities.

And the second one is potential new
markets like grid stabilization.

And those things.

So I think the use cases for AI
will only increase besides from the

original learn from your data idea.

John: It, it is, and it also.

Deals with some of the challenges around
staffing, which is you've got very,

you talked about the human knowledge.

Maybe I, I missed this point.

So the human knowledge, they're
deep expertise in in, in the

people who run these plants.

But just due to hiring cycles, a lot
of them are nearing retirement age.

And so there are, and I've met young,
bright engineers in these facilities,

but there aren't that many of them.

And it's gonna take them
a long time to learn.

the what, what these more experienced
people know and who's gonna teach

them if they've already retired.

So I think we have a, we, the other
thing that we have right now is an

opportunity to capture some of that
knowledge so that we don't lose it,

so we don't lose that knowledge.

That's that's, that's practical hands-on
knowledge of what to do when, and what,

what issues to, to, to look out for.

Denis: Yeah, absolutely.

I've, I'm really happy we came to this
conclusion because I was hoping at the

end of this podcast to have something
tangible of what we can do with ai.

Perhaps not what's being done
today, but something we are.

D growing towards, and I think
these applications are truly

exciting 'cause they truly give
a benefit to this industry.

John: Yeah, agreed.

We fully believe and, and this
kind of probably ties a bunch of

this together with that knowledge.

And multiple new sensors.

One new sensor has a, has a high chance
of getting stuck in that pilot state.

But if we bring multiple sensors
together so that the AI has more

to work with has, is giving bigger
insights and recommendations and

bigger responsibilities, that's
where the value really comes.

And that's, that's what we, that's where
we believe we will really be enabling

true AI solutions in the industry.

Denis: Mm-hmm.

So apparently I'm known for asking some
challenging questions at the end, but

given your experience, you have, you're,
you're working in a startup now serving.

Very big traditional companies.

Imagine, I imagine the listener
here could be someone working at

this big heavy company, perhaps
not even in touch with you, but is

a bit hesitant of where to start.

Let's say they have old
systems, old ways of thinking.

What in your opinion,
determines the success?

Criteria or probability of a
company that's going to make

it versus one that will fail.

What should companies start preparing
for right now, even if they're not so far

to make, let's say, this transformation
towards ai more likely to succeed?

John: So Aluminate, we talked to
senior, chief scientists, senior

engineers, directors, VPs at a
lot of aluminum companies and we

captured their knowledge to to a
solution that will help the industry.

And so if you're at that level, I
would, I would recommend getting in

touch with me and we will 'cause Kite
as a, as a whole, we are, we are very

interested in solving big problems
that are industry-wide, bringing AI

and sensors to solving those problems.

If you're someone who's in a facility,
in a plant who's, who's like this

ai, oh my God, I, I should use this.

There must be an opportunity there.

can I do?

I think it's, I, I think listening
to your podcast is a very

informative way of getting started.

and, and doing reading.

But you know.

For the success comes from the,
so what you are the expert.

These, if you're on the, if you're on
the floor, or if you are managing a

team on the floor, or if you're the
general manager of a facility that you

just, you feel like this could be done
better . What are your biggest problems

and what could a solution look like?

And I think if you can get to the
point of of identifying a significant

problem, you should be able to find
someone who's interested in, in tackling

that from from an AI point of view.

But for you, yourself, just as
a starting point, I think the

large language model application.

Bringing it in, starting using it.

just de-risks demystifies because it's
interesting what I, what I just said is

you create, identify a big problem and
then find someone to help you solve it.

I'm talking about it, I think,
well, that will work for some

people and, and, and not for others.

It's very easy to, how do you even
know what the AI is capable of today?

And so I think it's, if you can start
with low hanging fruit as well, that's

with bringing in low large language
models to help with your processes and

keep things running smoothly and taking
some of the, some of the routine stress

off people or having peop having the
routine activities occur more reliably.

That seems like a really
great place to start.

I dunno, Dennis, if you've
thought of other, other areas.

Denis: Well, yeah, I would just add from
my own area is that treat your data as

a valuable byproduct of your process.

And not just something at
your store in a database.

I think I used this quote before John
when I said that companies often assume

data is just lying around ready to
be used, but it's a bit like saying

that you have aluminum in the ground.

You can just dig it up with a shovel, but
no, to actually get it in a pure metal

form, you need to do a lot of processing.

And for some reason, when we think
about metals, we can visualize

this and understand this.

When we think about data, we
think it's already perfect and.

It's definitely not, you need
to do a lot of effort to improve

data quality and accessibility.

You can listen to our previous podcast
with Thomas about on this topic.

So take care of your data is
what I would add from my side.

John: I think that's, that,
that's a wonderful point.

And if, if, if you really want to
demystify the AI part, maybe just, for,

for someone who's sitting and, and not
in this world at all, but has some data.

If there's some data that you are
analyzing, if it, a time series

you're looking at, you're monitoring
temperature and you're looking for

excursions, maybe you're running SPC
on a, on a process and some, it's

plotting and you look at it every
day to see if it's in control or not.

Or there's start thinking about if,
if you're doing that and you have

a process that if you can generate
value for yourself from that data.

Then that's the time to start thinking
this would be a thing that's, that

would be a useful how can I automate
this using some of the AI tools.

and for me, AI is, and, and we
haven't dug into this as AI is more

of a workflow than anything else.

And and it, you can, you can
explore using standard workflow of

data plus analysis plus outcome.

And automating that, , you're
well on your way to understanding

how this can really help.

Denis: Yeah, absolutely.

I think that covers it quite well.

John, is there anything you would
like to add given this discussion?

Any afterthoughts.

John: We are at a very special
time, a point of transition where

there's a lot of value to be had.

If you look up the hype.

Cycle how the hype cycle works.

We are probably at at least one
peak on, on that hype cycle.

And so there will be a collapse when we
realize that not everything is possible.

Some of it will require persistence
and solving the same problems we've

had to solve before as people and
so we should be ready for that.

But I think certainly here at Aluminate,
we see a very promising future.

We see an industry that's, that's
under stress, doing what it can do

to respond to the Greenhouse Gas
Challenge and be a larger part of

a really important grid solution.

their willingness and openness
to for new sensors, new data.

New results that will help them.

So we see the future being
very bright in that industry.

And, and as we look around, there
are other industries as well

that where we're seeing similar
similar interest and enthusiasm.

And I, I think I think we will see we will
reach that that that value plateau of the

hype cycle sooner than, than we think.

Denis: Yeah, I have the same impression.

I think the last decade was more this
irrational exuberance about AI solving

everything, but with reaching a state
of maturity where we have solid products

solid companies delivering solutions.

So I think slowly and surely we're
going to get value from all of this.

John: Excellent.

Yes.

Thank you, Denis.

This was really enjoyable.

Denis: Yeah, it was great.

John, for the listeners that want to
learn more about you and Aluminate,

where can they find more about you?

John: Yeah, so Aluminate does
have a website, but we're it's,

it's a very high level concept.

We're not talking too detailed outside of
our conversations with our the partners

and the companies we're working with.

but if you want to contact me through
LinkedIn, I would be very happy

to hear from anyone and continue
or start a, a conversation there.

Denis: Great.

Great.

We'll put the links in
the show notes below.

Alright, John then, thank you very
much for being on this podcast.

and thank you also to the listener for
being with us on an other episode of

the Industrial Data Quality Podcast,
and I'll see you in the next one.

Bye-bye