How to Build an AI Advisor for Industrial Operations Using XMPro

🌟 Unlock the Full Potential of AI Advisors for Industrial Operations with XMPro

Welcome to this comprehensive webinar on configuring AI advisors using XMPro’s innovative data stream technology. As industries pivot towards more automated and intelligent systems, understanding how to effectively implement AI advisors is crucial for staying competitive. This session provides a detailed walkthrough of the XMPro platform, demonstrating the power of AI in enhancing strategic decision-making.

🔍 What You Will Learn:

AI Assistants vs. AI Advisors: Explore the distinction and advantages of AI assistants and advisors within your operational workflows.

Configuring AI Advisors with XMPro: Learn the step-by-step process of integrating advanced AI capabilities into your operations using XMPro's robust framework.

Boosting Strategic Insights: Discover how AI advisors offer more than just efficiency—they provide complex, high-level decision support that transforms how you approach business challenges.

🎯 Who Should Watch:

Industry Leaders: Operational Managers, CIOs, and CEOs looking to integrate AI to streamline decision processes and enhance operational efficiencies.

Technical Experts: IT Professionals, System Integrators, and Technologists interested in the specifics of AI implementation in industrial settings.

Innovators and Thought Leaders: Individuals driving digital transformation and interested in leveraging cutting-edge AI to solve real-world problems.

💡 Key Takeaways:

Actionable Insights: Gain a clear understanding of how to apply AI within XMPro to drive business outcomes.

Future of Operations: Learn about the shift towards decision automation and how XMPro supports this evolution with cutting-edge AI solutions.

Enhanced Decision-Making: See how XMPro’s AI advisors go beyond simple task execution to provide strategic, data-driven insights that are critical in today’s fast-paced market.

🚀 Join the AI Revolution in Industrial Operations

Embrace the future of operations with XMPro’s AI advisors. Whether you're looking to enhance operational efficiency, streamline decision-making processes, or adopt a more strategic approach to business challenges, this webinar is your gateway to understanding and applying generative AI in a meaningful way.

🔗 Stay Ahead of the Curve:

Subscribe to our channel for more insightful webinars and tutorials. Visit our website to explore more about XMPro and how our solutions are revolutionizing industries.

Connect with us on social media to join the conversation about the future of AI in industrial operations.

👍 Like, Subscribe, and Comment Below:

What challenges do you face in implementing AI in your operations? How can XMPro help you overcome these challenges? Share your thoughts and questions in the comments!

#XMPro #AIIntegration #GenerativeAI #DigitalTransformation #IndustrialAI #OperationalExcellence #StrategicDecisionMaking #AIWebinar

Transcript

good morning everybody and thank you for

joining me today so today what we're

going to do um in this webinar is run

through configuring an AI advisor using

data streams inside

XMR before we do that um and before we

jump in it'll be good for us just to get

an understanding of the the different

terminology that I'll be I'll be talking

to so what's the difference between a

assistant and a advisor those are the

key things that I'm looking to to run

through

here to do that um I'm going to break it

down from an aspect perspective we're

going to touch on assistant and advisor

uh side of things um as

well if we focus on the the aspect

pieces so breaking it down helps us to

compare the two of these things a little

bit easier so what is the focus of an

assist what's the focus of an advisor

what's their role approach some of the

benefits problem solving how do you

interact from a user perspective as well

as some of the value proposition uh

pieces as well so when we're talking

about an assistant its focus is

generally task and and operations um its

decision support capabilities are

immediate or

practical its expertise level is Broad

and general knowledge um so user

interactions generally direct or command

based um and and its main value

proposition why do you want an assistant

is efficiency and

convenience when we transition this to

an

advisor you'll notice that some of these

aspects start changing so again what's

the focus of an advisor is to give you

expertise and some strategy some of the

decision support there is high level and

complex one of the key benefits of a

advisor is strategic insights so we're

moving away from just increasing

productivity from an assistant to the

more insight side of things with an

advisor um the user interaction is a lot

more consultative so it's not so direct

or command based um it's more

interactive and you'll see the value

proposition right at the bottom there

when we start talking around an advisor

we're really talking around informed

decision making so an assistant is there

for efficiency and convenience uh when

we touch on an advisor we're looking a

lot more for informed decision making so

keep these in the back of the Mind as we

run through and and configure and look

at the different pieces of an AI visor

inside a and

BR if I break this down into so where

does this work with regards to uh what

we do at exent pop and there's

essentially three main areas of

decisions that can get made so you

generally start with decision support

that's your dashboards your your V by

condition monitoring typically Falls in

here if we looking at decision

augmentation this is where you are

reaching out and you're looking for

prescriptive recommendations augmented

information from an AI perspective or

from other systems uh Clos Loop feedback

Etc the last piece is around decision

automation where you have a human who is

not so much in the loop but on the loop

uh and keeping an eye on the automation

pieces um that are happening here the

current shift everyone is moving towards

decision

augmentation um and we see the future

shift moving towards decision automation

as

well so we're generally evolving and

moving from an informant to a performant

state from an XM Pro perspective we

cover the spectrum of all three so we go

from decision support to augmentation

all the way to decision automation um as

well what we're going to focus on today

though is we're just going to focus on

the decision augmentation piece this

piece in the middle over here um and and

how do we start bringing some of this AI

capability more specifically the AI

advisor pieces into this as

well how we see the the the AI and and

how it's working inside the operational

landscape is it's changing the landscape

but it's also helping um and move so

what do I mean by that so currently at

the moment uh you have automation so

deterministic workflow tasks you've also

got AI assistance so with the um chat

GPT um that's happened over the last

year and continuing into this year

everyone's very familiar with having a

conversation getting a response and

doing something that with that response

so AI assistant more freestyle chat side

of side of things as well um that

eventually if you remember the prior

visual that I had when we started

getting towards the more uh autonomous

on the far right you start moving into

the area of generative agents so how do

I have a goal-based single objective um

agent that will in turn move into how do

I have uh group goals so how do I have

an agent that looks off to specific

things and then how do I have a

supervisor role that actually helps

direct them from an agent perspective

and then the last piece is how do we get

to autonomous goal seing so

self-organizing um agents who don't

necessarily need a directed supervisory

role but can self-organize amongst

themselves and work out what they should

be doing and how they should be uh

interacting with the different systems

as

well you'll see on the far left here as

you increase your um automation

capabilities and your AI capabilities as

well you go up you start going into and

touching different areas of the outcome

pieces on the right here so when you

start looking at your task based genx

all the way up to uh self creating at

the

top how do we translate this into uh

what it is it that we do for industrial

operations so you'll see at the bottom I

still got the same items uh automation

assistant um I'll come to the advisor

agents and then multi-agent pieces as

well so I'll just build this up for it

so these are the different areas um that

you can actually interact with um inside

XM

Pro generic algorithms bring your own

models if you want to go towards the

more what we would call traditional AI

uh machine learning uh models and

algorithms um if you want to use some of

the notebooks we have done a prior

webinar on this as well so creating an

endtoend example

um using the accent Pro notebook so how

do I go from the one area all the way

through to to the other area um and make

use of the notebooks to to be able to do

that so that has been covered in a prior

webinar I encourage you to to go and

watch it if that is of Interest how do

you start bringing in some of the large

language pieces down here and this is

where we start transitioning into the AI

advisor pieces versus just an AI

assistant some of the other areas um

when we start talking around Ai and we

start talking around advisor is an

assistant the next thing that comes in

that is trying to do obviously rag or

retrieval augmented uh generation on top

of that so how can the advisors be a lot

smarter how can they be trained as

experts in what they are um to be able

to advise you correctly a key thing here

is also bringing your own llm model it's

one thing to be able to talk and go to a

chat GPT or anthropics clae but how can

you actually control the model that you

want to use and then on the far right

you'll see some of the agentic stuff um

we're not going to go into that as well

to today but just to give you a full

picture of all the different bits and

pieces from an exm Pro that you can

actually utilize uh for for your AI

Journey around industrial

operations before we jump into the

advisor if I just touch the AI assistant

so from an AI assistant perspective

there are a few different pieces in XM

Pro where you can use an assistant

the first is around

recommendations So currently the AI

assistant that you see here is in beta

um where you can have a discussion and

conversation with the assistant around

the

recommendations and what that will allow

you to do is to get some insights into

into the data as you triggered it um you

can ask it various questions it can

group sum it up um it can provide you

information from a graph perspective

that's tabular there's a lot of

different options in there but it is

just an assistant uh it just assists you

to find some of the information that

we're actually looking for the second

area where you can do that is inside

data streams now data streams are what

sits underneath everything that you're

looking at here there are data streams

behind the recommendations um that you

saw before so the AI assistant has data

stream sitting behind

it the data stream that you look at here

is using a open AI assistant agent now

what the open AI assistant agent does is

it allows you to interact and actually

talk to an assistant that you have

defined inside um open ai's uh

playground again what you can do there

is you can configure a specific set of

prompts you can spe create a specific

set of tools that it has access to and

then you can call it the challenge with

that though is you don't really have a

lot of control over fine-tuning the

prompt outside of that environment so

outside of that environment you have to

just use what was published which means

if you don't have access you can't

adjust the prompt you can't adjust the

model you're looking at you can't adjust

any of the parameters if that is your

use case and you don't need to then by

all means you can just drag the

component in from the toolbox and

actually just use it to talk to um the

open AI assistant from a cloud

perspective

there when we start getting to the

advisor

though there's a few elements U that we

are configured to bring this to uh to

life so to speak I'll run through them

here and then I'll actually take you

into an example for them as well so the

first is the result so where do you see

the AI ADV visor pieces um where can an

end user interact with them and how do

they visualize and see some of the data

for that as well the second piece is how

can you change and Define edit some of

the prompts that you're looking at you

want to be able to have fine grain

control over which model you're running

what are your prompts how can you change

them how can you shape them um as we're

going on the uh the AI Journey here and

interacting with large language models

prompting is one of the key elements to

make sure am I getting the right

information out that's relevant to what

I'm asking or is it just going to start

giving me incorrect information and

hallucinating the the other benefit of

creating advisor is you can create these

what we call experts in specific areas

so I can have an expert in rotating

equipment I can have an expert in

quality I can have an expert in safety

which means by narrowing their U scope

by narrowing their task Focus you help

to minimize a lot of the hallucinations

um that they can have as well if you

keep it too generic and too broad what

you will find is they will start

hallucinating just like we do um and the

results are not going to be what you're

expecting the third element to this is

we're actually using

recommendations so we created a generic

recommendation that we are tying to the

data streams that allow me to push it

and make it available to the first

screen that you saw here which was the

actual um an app and last but what by

definitely means uh most important piece

here is how do I actually create this so

data stream underpin everything that

we're looking at so that is the

information of where I get my data run

my models and actually present it to the

end users on the other side so let's

jump into an example and configure what

I'm looking at over here so the first

piece of the puzzle is let's have a look

at the data stream and then I'll work

backwards from

there when you look at this particular

data stream I'll draw your attention

right down here to the actual um the AI

piece and what it's actually doing so

this stream when it's running is getting

data coming from an endpoint now again

this endpoint can be anything there are

a lot of different listeners in here

from a library perspective so all you

have to do is drag it on if you don't

want to use mqt you want to use opcua or

a historian by all means you can do that

as well you'll see we're getting the

prompt details in here so we're reading

the prompt details from a data source

this is coming out of SQL can it come

out of a graph database yes can to come

out of another system yes you want to

store your prompts in CSV flat files and

read that in you can do that as well the

key piece here is you want to be able to

have a generic mechanism to get your

prompt Details

In The Stream here is also using

recommendation so if I have triggered a

recommendation for this particular asset

do I want to re-trigger one or do I want

to even pass the information to the

model and have it run no I don't the key

thing to remember here

is you don't want to push every single

thing through at a speed of you know

every 1 second to a model because models

do take some time to run if you are

running this model in the cloud and you

want to get the per second you can

achieve that yes it's going to be costly

if you want to run this model U locally

again you can achieve that but it's

going to be costly in infrastructure

costs the reality of the use case here

is to work out under what condition do I

actually want to pass this information

to an advisor and get a response back so

what you'll see here is we've actually

got a filter that says if there's radio

recommendation for this asset that was

triggered by this AI advisor we actually

don't want to pass anything to the model

we just ignore it because there's

already a recommendation um that's been

triggered if a recommendation has not

been triggered then we're going to pass

it to the actual model that we're

looking at here it's going to evaluate

what we're giving it and it's going to

give us the information and we are then

going to run a recommendation and update

the

recommendation if I double click any of

these there's nothing weirdly

complicated about this the actual

configuration will just walk you through

what you're actually looking at here

it's the exact same data streams uh for

those who have used these in the past

that you are familiar with all that

we've got access to here is a new agent

in the machine learning side of things

so you'll see here is an AER open AI

I'll scroll a bit further down here is

aama Agent there is the open AI

assistant so all there are are new

agents in the actual library that you

can just use in the same way that you

are used to um in the

past if I double click the SQL that I've

got here you can see we're looking at

the prompt if I double click this here

this is actually pointing to a local

model you have a few different options

when it comes to U how do I want to

configure an AI advisor as I mentioned

earlier I can go and push this to an AI

assistant um in open AI who you can

train as an expert get the results that

come back and then add all the different

other recommendations um around that and

then you can get a AI advisor piece

coming out of that the challenge with

talking to to the cloud ones is there is

a cost around the

tokens so it does not matter how many

tokens you send there is a cost to

compute that and get the results for

that to come back the other option is

you can run models locally so this is

actually running locally in uh the lab

that I have set up here and I can pass

the URL on the model in dynamically I

can also pass all the um system prompt

and the user prompts in dynamically as

well if I want to change out it's

actually pretty quick and easy for me to

do that if I want to adjust for instance

model temperature um I can adjust that

as

well when the stream is

running the Telemetry details coming in

here we'll send our information through

what you'll notice here is I have a

pre-filter so what I'm looking for is

I'm only looking for assets that have

gone above over and above or below

certain thresholds that I've to find I

don't want to use the llm to actually

filter out noise I actually want to

filter out the noise before I get to

should I pass this to a um large

language model for it to evaluate

further um otherwise you're are going to

just be wasting resources down here and

passing it noise and hoping it figures

out some of the pieces as well can you

do that you can again you're going to

have to take into account token costs if

this is going into the cloud or you're

going to have to take into account

compute costs from a infrastructure

perspective locally if you're running a

local one

so this is what sits underneath the hood

data is coming in we are getting some

prompts we're also using the

recommendations to help weed out before

we get to run the actual model

itself the result where do I actually

see the result of of this if I look at a

particular app that I've got over here

it's the same app that we used to and

and have been looking at in the past on

the right you'll see here are the

recommendations if I click the top

recommend Commendation this will open up

a recommendation that was triggered by

the AI advisor so this is the result of

that data stream that I'm looking at

here you'll see in the middle piece here

AI response this is what the actual

large language model gave me when it ran

through this particular data stream and

triggered a recommendation

here so again this is how an end user

can interact with it if we go back one

more step so I've got a data

stream the data stream was actually

triggering a

recommendation so if we go into well

what does that recommendation look like

exactly the same as you're used to we

have our recommendations you will see

they are configured and set to a

particular data stream so it's exactly

the same data stream that we are looking

at

here and we have created a generic um

recommendation for it you still got

access to all the detail and value so I

can be a lot more specific in the

headline and descriptions with the data

that is passing through the data stream

as well so exactly the same behavior you

used to when you're configuring

recommendations for other use cases this

is just using it from a generic

perspective to create an AI advisor that

I can use on the apps

themselves if we go back to the

app this is a recommendation if I go

back one you will see it'll take me back

to the um application itself and I've

got access to all my other

recommendations now for those who are

familiar with some recommendations this

may look a little bit

different this is a recommendation uh

with a different template that was

applied to it you can still use the

prior recommendations and if that is um

where you want to put all of this but

you don't have to you can adjust the

visuals to other types of visualizations

for the same set of data so here I still

have my event data as you used to I'll

scroll a bit further down I'll still

have my analytics I can still have my

notes coming in here as well across the

top you can assign it false positive or

resolve the recommendation as well so

exactly the same capability present it a

little bit

differently this is the recommendation

that triggered for it if we go back to

the stream that gets used down

here the recommendation that is

triggered gets read in here which stops

me from talking to the model too many

times let's have a look at the prompt

over here though so if I have a look at

the prompt you'll see it's coming from

SQL there is a actual application that

you can configure um on top of that as

well if you already have a prompt

Library um and you just want to read

that in we can do that as well this one

pretty simple and straightforward what

model am I looking to run what's my

model temperature what is my system

context for that and my user

context system context defines that you

are an expert in whatever for this

example it's rotating equipment so he's

an expert in the pump efficiency and

optimization you can create an expert as

I mentioned earlier from a safety

perspective from a quality perspective

maybe there's some static equipment or

different types of rotating equipment

you can also create an expert in the

process flow so let's take a

manufacturing example you can create an

expert

and the actual flow all the way from

start of manufacturing to the end of

manufacturing as well user content is as

you are used to when you are creating

your own uh conversations with your gpts

uh which pieces do we actually want to

plug

in when we are running this through the

Stream So when we say user content here

what we're actually meaning is in this

instance the user content will actually

be coming from whatever source of data

it is there's no actual user um as a

human that is sitting and talking to

this your machines are actually talking

to it in this example can this read from

a human yes it can if you want to

actually expand the um AI assistant that

I mentioned earlier with the

recommendations and actually have a

human ask similar questions you can do

that as

well so we're getting in the prompt

details and if you have a look at how

they have been defined II assistant with

an expertise what is your top askask

that you are looking to do and what

should you be

considering depending on which model I

want to talk to I may want to change and

tune my system prompts I might want to

change and tune my user prompts I can do

that quite easily over here save the

changes and whatever the interval is

that we've defined here so if I go back

in here and have a look every five uh

minutes it will get the new prompt if I

want to change this and say you know

what I'm actually trying to split test

some prompts I can change that to every

5 Seconds as an example which means if I

save my new prompt details away I just

need to wait 5 seconds and then it'll

load in the the new prompts and I can

then evaluate that I don't need to stop

and start a stream I don't need to stop

and start and publish anything I can use

an interface like this just to update

and change what I'm looking at the

benefit of running

a uh a local offline um Repository like

olama is I could have one prompt that is

using an olama you'll see I have another

prompt here using SQL coder as a model

Etc so I can have different models

running for different types of assets

for different use cases all running

within the same structure that I've got

defined

here so this is the prompt that is

sitting behind and it's being fed into

the data stream this is the

recommendation that we've configured and

set up that I can interact with and from

an end user this is how they can consume

and react to the responses coming in now

some key things here is it's always good

to make sure that you have a mechanism

to give feedback to the model as well is

this a good response or a bad response

um for it so it can also help learn that

the next time it runs through it takes

that into account as

well if I'm looking at this and I

actually want to have a conversation to

it can I do that as well yes you can

you'll see on the right here I can start

having a conversation and it can take

into account this response it can take

into account the event data

and and my horrible spelling as well so

information in the middle how we got

that that we got that through the data

stream triggering that recommendation

there this model was actually coming off

of a local

model where over here as this is running

this is actually talking to a um chat

GPT I think this is running chat for all

and I can have conversations with it

with different data again prompts as you

can move these around the data as you

can move them around all we're doing

here is we're just putting the different

pieces together

around a use case um that we call as a

an AI advisor on the data that is coming

in some of the information that you

looking at in here as well you can see

some of the citations coming in as well

so again grounded in where's my

information coming from so that I can

trust the data as

well so what do we just go through from

an AI visor perspective there's a few

different elements to to consider here

the first is the

visual where can I consume the data can

we pass this AI response to other

systems yes we can this is just an

example inside X Ando we can pass this

to any Downstream system uh that

requires it needs it or once

it the next piece is the actual prompts

themselves how do I create the prompts

change prompts maybe I want to do some

prompt chaining in here how do I tune

the prompts for the situation ation that

I'm creating or want to uh help

manage if we go one step further the

generic recommendation so here we're

just using our recommendation engine and

creating a generic recommendation for

the AI side can you be more specific and

actually create recommendations for

specific asset classes yes you can can

you hook this into existing data streams

that you may have to finded you can do

that as well this does not have to stand

on its

it fits in with everything that you used

to um and have configured up to now as

well and then the last is the data

Stream So what sits behind this what is

actually consuming the data bringing the

prompts in Reading recommendations

passing it to the model getting the

result and then passing that to the

um the recommendation so that they can

be triggered as well the last side I'll

leave you with is again what are all the

different pieces when it comes to to the

AI side of things for industrial

operations there's generic

algorithms you can bring your own model

if you want to do just simple regression

Etc there is notebook capability and as

I mentioned earlier there is a webinar

that we have done previously which will

walk you end to end through that as

well this is where we start getting into

some of the large language models so

there are agents for for

instance um o llama to run it locally

there are some to run open AI um and

that library is expanding as more models

become available on the far right when

we start dealing with the advisor side

of things how do we do that on top of

xmo recommendations how do we also start

doing that on top of your own

documentation as well so how can we

perform rag which again is retrieval

augmented generation on top of your own

data so we want to use your own data

your own manuals Etc as part of that

advisory uh

capability and then bring your own model

maybe you've defined your own model and

you don't want to use one of the

available ones out there you can bring

that in as well and just some of the

other future work that we are working on

right now is around the agent side of

things so how do you create a generative

agent a directive agent and also moving

towards multi-agent systems where they

are self-organizing as

well and with that I thank you for the

uh for the time today thank you for

listening to today's webinar I hope you

have a great rest of uh rest of the day

and we'll see you again on one of our

future webinars thank you