Grounded AI: Making Agentic AI Work for Complex Engineering Systems

WEBVTT
00:04.867 --> 00:09.091
Hello
00:09.318 --> 00:12.813
there, everyone. Welcome to episode
00:13.313 --> 00:17.715
number 693 of this here electronic
00:17.796 --> 00:21.977
engineering podcast called Amelia's Weekly
00:22.090 --> 00:26.811
Fish Fry, brought to you by eejournal.com
00:26.924 --> 00:29.866
and written, produced, and hosted by yours truly, Amelia Dalton.
00:32.813 --> 00:37.203
A lot of folks are talking about agentic AI like it's some kind
00:37.703 --> 00:39.815
of magic shortcut for engineering,
00:40.423 --> 00:43.868
promising lightning-fast automation and effortless
00:44.368 --> 00:48.162
results. But in the world of high-stakes systems,
00:48.691 --> 00:52.008
speed doesn't mean much if you can't trust the output.
00:52.825 --> 00:56.446
If your AI is doing its own thing, disconnected from
00:56.946 --> 01:00.019
the simulation and verification we all rely on,
01:00.998 --> 01:04.227
How can you really ever trust it? My guest
01:04.727 --> 01:08.790
today is Jason Guidella, Senior Principal Technologist
01:08.951 --> 01:13.272
at MathWorks. And we're moving past this going
01:13.385 --> 01:17.048
faster hype to look at the actual challenges
01:17.305 --> 01:21.064
of deployment. Jason and I discuss what
01:21.177 --> 01:25.691
breaks when you try to use agentic AI in complex
01:25.755 --> 01:29.846
engineering systems. Where today's agentic AI
01:30.346 --> 01:33.100
tools tend to fall short for engineering teams,
01:33.774 --> 01:37.461
and why grounded AI, the kind that actually works
01:37.961 --> 01:41.692
within your existing engineering workflows, is the
01:41.773 --> 01:45.027
only way to go. So without further ado,
01:45.395 --> 01:48.762
please welcome Jason to Fish Fry. Hi,
01:48.842 --> 01:52.048
Jason. Thank you so much for joining me. Thanks,
01:52.064 --> 01:55.337
Amelia. Great to be here with you. Excellent. Okay,
01:55.433 --> 01:59.450
so Jason, everyone talks about how much faster agentic
01:59.579 --> 02:03.499
AI can make engineers, but is speed
02:03.596 --> 02:07.597
actually the hard part? Speed is what gets the attention,
02:07.950 --> 02:10.328
and it's an important part of the overall story.
02:11.308 --> 02:14.618
However, agentic AI can create code,
02:14.683 --> 02:17.880
it can modify models, it can set up simulations,
02:17.977 --> 02:21.463
tune parameters, it can do so much. But that's not what
02:21.963 --> 02:25.173
engineers need only. Engineering is much more
02:25.237 --> 02:28.433
than just that. And that's not the hard part. The harder question
02:28.933 --> 02:32.689
is whether that work that is being done can be evaluated with inside
02:33.189 --> 02:36.623
an engineering workflow. So this is where I think we need
02:37.123 --> 02:40.317
to be clear on what the role of the agent is. The agent can help
02:40.817 --> 02:44.734
create a model. It can, it can propose a change, create an app, maybe create
02:45.234 --> 02:48.762
test cases and orchestrate a workflow. But it shouldn't be replacing
02:48.906 --> 02:52.290
deterministic engineering tools. It should not make
02:52.790 --> 02:56.283
up a simulation result. It should not replace static code analysis.
02:57.004 --> 03:00.693
It should call the right deterministic tools with inside
03:01.193 --> 03:05.071
our engineering workflow to do those things. So I think the real reason
03:05.087 --> 03:08.246
here is simple. Engineering evidence has to be repeatable.
03:08.406 --> 03:11.982
If I give the same model, the same code, the same requirements,
03:11.998 --> 03:15.092
and the same test inputs, to an analysis workload, it needs
03:15.124 --> 03:18.698
to give me the same result back. And that determinism is
03:18.810 --> 03:22.560
what allows engineers to use the evidence in
03:23.060 --> 03:26.374
a decision process. So a concrete example that I
03:26.823 --> 03:30.412
like to use, it's a search and rescue system that's called
03:30.912 --> 03:34.611
LifeSeeker, and it's made by Sentum. And it's at one level, you can
03:34.739 --> 03:37.607
really think of this system, it's a very easy system to understand.
03:37.736 --> 03:41.022
Someone is lost maybe in the Alps. A search team is trying
03:41.522 --> 03:45.222
to locate them, and the system helps find them really a lot,
03:45.270 --> 03:49.069
much faster. Now, from an engineering perspective,
03:49.502 --> 03:53.157
that's a highly interconnected system. It's involving RF signals,
03:53.269 --> 03:55.561
signal processing, communications,
03:55.754 --> 03:59.168
aircraft motion, geolocation algorithms,
03:59.281 --> 04:02.567
embedded software, and operator workflows. And so this
04:03.067 --> 04:06.655
is a really complex system. And you want an
04:07.155 --> 04:10.710
agentic workflow here to help you develop it, but you only
04:10.822 --> 04:15.038
want it to do that work if it's within the way engineers
04:15.391 --> 04:18.757
are building and evaluating that full system. So the way I like to
04:19.257 --> 04:22.893
frame it is that the bottleneck is shifting. Before,
04:22.909 --> 04:26.099
the first question was that the— can the AI create the engineering
04:26.195 --> 04:29.449
work? And I think that's increasingly becoming, yes, it can.
04:30.235 --> 04:33.906
So the next question is, can that work be evaluated in the
04:34.406 --> 04:37.401
same workflow that engineers already use to make design decisions.
04:38.572 --> 04:42.807
So, so like, if I take that LifeCycle Seeker example
04:43.307 --> 04:47.187
a little further, so let's suppose the agent has proposed a new improvement
04:47.685 --> 04:51.872
to the signal processing algorithm that's being used to locate a lost
04:52.372 --> 04:55.915
signal, cellular signal. Now, that sounds valuable, and it
04:55.947 --> 04:59.621
should be evaluated. However, the engineering team has
04:59.653 --> 05:02.685
to do that evaluation across the whole system.
05:03.086 --> 05:06.595
Does it work under a weak signal condition? Does it change system latency?
05:07.108 --> 05:10.586
What about the computational effects? These are all things that need
05:11.355 --> 05:14.848
to be resolved. It can't be just simply, does it work better?
05:15.361 --> 05:18.293
What I needed to do is I wanted to run the simulation,
05:18.486 --> 05:22.460
execute the relevant tests, check the analysis, and preserve
05:22.492 --> 05:26.081
that traceability needed to understand if this change is happening.
05:26.434 --> 05:30.234
So for me, the hard part here is It's not simply creating
05:30.347 --> 05:33.991
engineering output faster, it's creating output that fits into
05:34.023 --> 05:36.848
the workflow engineers use to generate deterministic,
05:37.313 --> 05:39.866
repeatable evidence here. So, Jason,
05:40.219 --> 05:43.718
what's different about using agentic AI in engineering
05:44.216 --> 05:47.378
versus using it to write code or documents?
05:47.924 --> 05:51.664
So I think with the difference here is with engineering work, it's tied
05:51.777 --> 05:55.341
to system behavior, and engineering evidence has to be
05:55.373 --> 05:58.820
repeatable here. So when I look at AI to help me write a
05:59.320 --> 06:02.956
document, a person can quickly review whether the language communicates
06:03.004 --> 06:06.643
what you intended it to say. With AI,
06:07.172 --> 06:10.587
if it can help you create code, that code
06:10.683 --> 06:13.905
has to still be tested. And in an embedded system
06:14.370 --> 06:17.111
like a signal processing communication control system,
06:17.897 --> 06:22.097
that chain really happens and stays local to that particular component.
06:22.803 --> 06:26.354
And so that software change can affect timing. It can
06:26.854 --> 06:30.772
affect the detection performance. It could change the actual way the physical system
06:30.836 --> 06:34.210
reacts, right? So all of these things need to be
06:34.710 --> 06:38.870
taken into consideration. And so the engineering question is not simply, did the AI
06:39.370 --> 06:43.384
create something that looks right? It's really that, how does that change behave
06:43.448 --> 06:46.822
in the system? And what evidence do we have to prove that? So I
06:47.322 --> 06:50.419
think this is where the agent role matters. I want the agent to help me
06:50.919 --> 06:54.572
create a model, create the supporting code, maybe suggest test cases
06:54.941 --> 06:58.085
and orchestrate, say, an analysis sequence here.
06:58.838 --> 07:03.152
But when it comes to evidence, the agent should be calling established engineering
07:03.184 --> 07:07.434
tools, simulation, verification, code generation, static code analysis,
07:07.771 --> 07:11.251
test frameworks, and such. You know, we've talked about
07:11.508 --> 07:15.164
this, I think, and you too, Amelia, have talked about that with my colleagues over
07:15.212 --> 07:18.789
the past few months, digital twins, model-based system engineering,
07:19.366 --> 07:22.803
RF and software-defined systems. And those conversations really
07:22.915 --> 07:26.497
were all about a similar flow, system-level visibility,
07:26.673 --> 07:29.789
virtual validation, simulation before moving
07:29.821 --> 07:32.952
to hardware, and then reducing that integration risk.
07:33.755 --> 07:37.818
And so that's, that's the important piece here, I think, is that the agent can
07:37.834 --> 07:41.560
help only when it's inside and participates
07:41.608 --> 07:44.997
within the engineering workflow rather than operating as a separate
07:45.497 --> 07:48.880
source of truth. All right. So Jason, why are
07:48.928 --> 07:52.345
engineering workflows built around simulation,
07:52.521 --> 07:55.794
verification, and traceability in the first place?
07:56.468 --> 08:00.205
I think it really comes down to engineers need objective evidence.
08:00.333 --> 08:03.221
So simulation gives evidence about the behavior.
08:03.606 --> 08:07.696
Verification gives evidence that requirements or design expectations
08:08.196 --> 08:11.851
are being met. Testing gives evidence that the change
08:11.867 --> 08:15.843
works under defined conditions. And you need traceability to
08:15.859 --> 08:20.010
connect that original intent from the design to the implementation
08:20.202 --> 08:24.017
and to the results that were used to evaluate it. But there's also another important
08:24.209 --> 08:27.270
point here. This evidence has to be deterministic
08:27.559 --> 08:31.341
and reproducible. If the same model and same test conditions produce
08:31.841 --> 08:35.476
different conclusions every time, that's not a basis for any engineering
08:35.492 --> 08:39.259
decision-making, right? And this is why existing engineering tools do
08:39.275 --> 08:42.576
matter. I need a simulation engine. I need a code generator
08:42.624 --> 08:46.891
or a study. Analysis tool or a test framework here because they
08:46.955 --> 08:50.549
operate within defined inputs, defined configurations,
08:50.565 --> 08:53.966
and produce repeatable outputs. And those tools make
08:54.030 --> 08:56.404
that needed evidence that engineers have.
08:57.302 --> 09:00.703
When you look at it like an agentic AI can help
09:01.203 --> 09:04.537
set up the workflow, it can help identify what needs to be checked,
09:04.778 --> 09:07.922
it can help create the artifacts or call the next step.
09:08.628 --> 09:11.885
But it should be calling the deterministic tools to produce the evidence
09:12.077 --> 09:15.849
and it shouldn't be making that up. And why workflows need to exist
09:16.349 --> 09:20.216
in the first place is that the team needs this repeatable evidence
09:20.650 --> 09:24.295
that the system behaves as expected, not just as an AI
09:24.359 --> 09:27.377
explanation that sounds plausible in the first place.
09:28.373 --> 09:31.664
And, and I think when you look at it, this connects
09:32.164 --> 09:35.951
back to our core ideas with model-based design and even model-based system
09:36.000 --> 09:40.255
engineering, that models are not just documentation. They are executable
09:40.755 --> 09:44.731
representations that help team evaluate behavior earlier,
09:44.827 --> 09:48.885
it helps you explore alternatives and understand the effects of changes
09:48.917 --> 09:52.236
before deployment. Absolutely. So Jason,
09:52.445 --> 09:55.524
what breaks when you try to use agentic AI in
09:55.636 --> 09:59.068
a complex engineering system? So I think what
09:59.084 --> 10:03.558
breaks is often the connection between what the AI-generated
10:03.734 --> 10:07.439
output was made and the engineering evidence needed to
10:07.939 --> 10:11.798
evaluate that output. So let's stay with the LifeSeker
10:11.991 --> 10:16.061
customer example. So suppose an agent proposes
10:16.173 --> 10:20.067
a change to improve that signal detection or location accuracy.
10:20.660 --> 10:23.881
This is going to be a very useful change to have. But now the engineer
10:24.381 --> 10:28.015
needs traceability and impact analysis to understand what
10:28.515 --> 10:31.557
requirement does this change relate to? What part of the model
10:32.057 --> 10:36.348
changed? What tests should be rerun? You know, what downstream
10:36.364 --> 10:39.456
behavior might be affected and so on.
10:40.129 --> 10:43.686
This is also where determinism matters. The agent
10:44.215 --> 10:47.547
should not answer those questions by inventing its own analysis.
10:48.140 --> 10:51.393
It should call the tools the engineers use for those tasks.
10:51.601 --> 10:54.341
It should run the simulation. It should invoke the tests.
10:55.030 --> 10:58.282
It should use the static analysis tools. It should preserve
10:58.314 --> 11:01.727
the links to requirements and results so engineers can inspect
11:01.839 --> 11:05.370
the impact of what might happen. And if the agent makes
11:05.434 --> 11:09.209
the change outside of that workflow, engineer has to reconstruct
11:09.290 --> 11:12.728
all of this manually. They need to inspect the change,
11:12.824 --> 11:16.536
map it back to the requirements, determine what simulation and tests are
11:17.036 --> 11:20.231
relevant, run the analysis and compare results. And so all that
11:20.731 --> 11:24.537
productivity that you've, you've gained, you've just now lost it on the verification
11:24.649 --> 11:29.244
side. And this issue is not just whether AI produced
11:29.744 --> 11:33.900
a good answer. The issue is whether the workflow is preserved, traceability, the impact
11:34.400 --> 11:38.138
analysis, and the repeatable evidence needed to really evaluate that answer.
11:38.764 --> 11:42.906
So kind of following up on that, Jason, why isn't it enough
11:43.098 --> 11:46.790
for agentic AI to just give engineers the right
11:47.290 --> 11:50.755
answer? Because in engineering, the right answer without supporting evidence
11:51.255 --> 11:54.720
is still incomplete, right? It's not an answer. I like to say sometimes,
11:54.977 --> 11:58.525
show me the money. Don't just talk about it. I'll eat everything.
11:58.669 --> 12:01.865
Show me the money. Show me the evidence. But really, more seriously,
12:02.057 --> 12:05.846
engineers need to understand what changed, what assumptions were made,
12:05.911 --> 12:08.929
what simulations were run, what tests passed or failed,
12:09.363 --> 12:12.413
and whether the result is traceable back to that design intent.
12:13.280 --> 12:16.235
That's especially important in embedded signal processing,
12:16.331 --> 12:18.884
communication, and control systems. In these systems,
12:19.703 --> 12:22.914
the behavior comes from the interactions between the algorithm, the software,
12:22.994 --> 12:26.029
the hardware, the physical dynamics, sensors, actuators,
12:26.542 --> 12:30.089
communication links, the real-world operating conditions. So if an
12:30.137 --> 12:32.770
agent proposes a new signal processing method,
12:33.604 --> 12:37.215
I do not only want that method, I want to know how it
12:37.715 --> 12:41.533
performs across representative signal conditions. So I want to know whether it affects
12:42.033 --> 12:45.641
latency or resource usage, right? I want to know which test cases were
12:46.141 --> 12:48.803
run. And I want to know whether it still satisfies my requirements.
12:49.413 --> 12:52.494
And I want those results to come from
12:52.527 --> 12:56.411
the tools engineers already depend on. If the question is whether generated
12:56.796 --> 13:00.536
code matches the model, use the code generation and verification
13:00.648 --> 13:04.500
workflow. If the question is whether code has defects,
13:04.741 --> 13:08.721
run the static code analysis. If the question is how does the system
13:08.753 --> 13:12.188
behave, run the simulation. And the question is whether
13:12.688 --> 13:16.521
the requirements are still being met, check the traceability and test results, right? And this
13:17.021 --> 13:21.401
is where really MATLAB and Simulink are directly relevant. Engineers have used MATLAB
13:21.901 --> 13:25.305
and Simulink to model systems, simulate behavior, analyze results,
13:25.434 --> 13:28.406
create tests, and connect requirements,
13:28.422 --> 13:31.780
generate code, all these things, right, to support that overall
13:32.117 --> 13:35.652
verification workflow. And for agentic tools to
13:36.152 --> 13:39.878
be useful in that environment, those tools need access to that same engineering
13:39.910 --> 13:43.188
context. It needs to work with the
13:43.688 --> 13:47.349
models, the simulations, the code, the tests, and workflows engineers already use.
13:48.103 --> 13:51.774
And so that's why we made— released the MATLAB MCP
13:52.274 --> 13:55.813
Core Server and the MATLAB Agentic Toolkit and Simulink Agentic Toolkits.
13:56.518 --> 14:00.366
They're not replacing MATLAB and Simulink with an AI agent, but they're
14:00.866 --> 14:04.021
giving the agent a structured way to use MATLAB and Simulink effectively,
14:04.453 --> 14:07.387
to call the right tools, inspect artifacts, run analyses,
14:07.740 --> 14:11.747
and move work through the workflow. So the goal is not for AI that
14:12.247 --> 14:16.139
simply gives an engineer an answer. The goal is that the AI is helping create
14:16.639 --> 14:20.694
the answer and use the right deterministic engineering tools to produce the evidence needed
14:21.047 --> 14:24.241
to evaluate it. So Jason, where do
14:24.741 --> 14:28.383
you think today's agentic AI tools tend to fall short
14:28.495 --> 14:31.898
for engineering teams? Many agentic AI
14:32.043 --> 14:35.654
tools can create content or automate tasks,
14:35.815 --> 14:39.362
but engineering tools need more than task automation.
14:40.053 --> 14:43.359
An agent has to do more than create the code or suggest a model change.
14:43.761 --> 14:47.661
They are very useful activities and capabilities that will help speed,
14:48.015 --> 14:51.788
right, as we mentioned. However, if the agent is disconnected from
14:51.852 --> 14:55.449
the engineering environment, it may not have the right context
14:55.561 --> 14:58.773
or skills needed to work in a way that engineers can use it.
14:59.575 --> 15:03.188
It may not understand the system model. It may not know what the requirements
15:03.688 --> 15:07.026
are. It may not know which simulations matter. It may not know
15:07.106 --> 15:10.638
which tests are relevant, or it may not even know how a
15:11.138 --> 15:13.946
change affects other parts of the system. And these are, you know,
15:14.122 --> 15:17.541
so important. And I think on top of that, if it doesn't know
15:17.621 --> 15:21.297
which deterministic tool to use to produce the evidence needed,
15:21.473 --> 15:25.197
then again, you're, you're broken from this overall workflow. If we go
15:25.261 --> 15:28.696
back to our Life Seeker example, if the agent is used
15:29.196 --> 15:31.167
to create that signal processing algorithm,
15:32.483 --> 15:36.416
that's useful, but its usefulness only depends on much more
15:36.916 --> 15:40.621
than just how it creates that code. The team needs to know if that
15:40.669 --> 15:44.361
code behaves in the system. They need to test it against
15:44.861 --> 15:48.438
the signal conditions and geolocation performance and embedded
15:48.938 --> 15:51.985
constraints and timing constraints and system-level requirements.
15:52.547 --> 15:56.576
This is where I would distinguish between generic automation
15:57.076 --> 16:00.476
and engineering automation. I would say a generic automation is that the
16:00.492 --> 16:04.264
agent completed a task, and the engineering automation is that
16:04.764 --> 16:08.198
the agent completed that task in a way that it is connected to the
16:08.698 --> 16:12.744
workflow used to evaluate that task. That is why that context and skills
16:13.178 --> 16:16.631
matter, and that's why agents need the right context
16:17.131 --> 16:21.016
from the engineering environment that they're working within. And that's the way they can
16:21.113 --> 16:24.405
work most efficiently and correctly and have access to the
16:24.905 --> 16:28.742
right tools. So for MATLAB and Simulink, that means structured
16:29.242 --> 16:33.141
access to the models, the simulations, the tests, and so on. And that's where the
16:33.641 --> 16:37.900
MATLAB MCP Core Server and the Agentic Toolkits help. They help agents
16:38.044 --> 16:41.345
operate through engineering tools and workflows rather than around them.
16:41.761 --> 16:44.582
So Jason, I've heard the term grounded AI.
16:44.806 --> 16:48.427
So talk to me about this. When I
16:48.619 --> 16:52.625
hear that term, when I use that term, I mean AI that
16:52.705 --> 16:56.807
is operating within that engineering context that engineers already use
16:57.307 --> 17:01.134
to design, simulate, test, verify, and deploy systems. It is grounded
17:01.294 --> 17:04.677
in the models. It is grounded in the requirements. It is grounded in
17:05.177 --> 17:09.229
simulation results, test cases, verification workflows, engineering constraints,
17:09.309 --> 17:12.692
and so on. I think also grounded AI for me also
17:12.756 --> 17:16.491
means respecting the boundary between what an AI should
17:16.619 --> 17:19.825
do and what deterministic engineering tools should do.
17:20.595 --> 17:24.314
The agent can help again, create a model. It can propose a
17:24.362 --> 17:28.083
design change. It can create an app. Suggest the right test cases
17:28.131 --> 17:31.439
to make, and it can orchestrate workflow steps for me.
17:32.097 --> 17:35.533
It could also help explain the results and, and help engineers navigate the
17:36.033 --> 17:39.675
complexity of an overall design. But when the workflow requires
17:39.787 --> 17:43.159
engineering evidence, the agent should call the appropriate tool.
17:43.560 --> 17:47.638
It should run the simulation, it should invoke that test, it should use
17:48.138 --> 17:50.994
the right code generator, run the static analysis, and so on.
17:51.588 --> 17:54.956
The agent should not become a probabilistic replacement
17:55.456 --> 17:59.300
for their steps, right? And I think this is a really important distinction because engineering
17:59.364 --> 18:02.763
evidence has to be repeatable. Same input, same configuration,
18:02.811 --> 18:06.306
same tool, same result. That determinism is
18:06.402 --> 18:09.464
what makes engineers make decisions from that evidence.
18:10.105 --> 18:13.888
I think this is where MATLAB and Simulink and the MATLAB MPC Core Server,
18:13.937 --> 18:17.672
the MATLAB Agentic Toolkit, the MATLAB and the Simulink Agentic Toolkit
18:17.784 --> 18:21.587
come together. MATLAB and Simulink provide that engineering environment.
18:21.876 --> 18:25.600
The core server and agentic toolkits give the agents that
18:26.100 --> 18:30.063
structured way to access those capabilities and use them as part
18:30.143 --> 18:33.369
of an overall engineering workflow. I'll come back to my Life
18:33.869 --> 18:37.575
Seeker example. The grounded AI would not just propose a
18:37.623 --> 18:41.588
better algorithm, it would help make the proposal in the
18:41.700 --> 18:45.247
engineering environment where the team can evaluate its effect on
18:45.747 --> 18:49.422
the RF behavior, the signal processing, the embedded constraints,
18:49.503 --> 18:53.036
the timing, and system-level requirements. And that is the shift
18:53.536 --> 18:56.955
I think matters. Agentic AI becomes more useful when
18:57.455 --> 19:00.986
it does not bypass the engineering workflow, but participates in it.
19:02.255 --> 19:06.398
All right, Jason, before I let you go, it's time for your off-the-cuff question.
19:06.944 --> 19:10.221
So if you could have one meal right now, it doesn't
19:10.721 --> 19:13.673
matter if it's on the other side of the world, you need a passport to
19:14.173 --> 19:18.327
get there, what would you have? I would have Moreton Bay bugs on
19:18.360 --> 19:21.641
the Esplanade in Cairns in Far North Queensland where
19:21.674 --> 19:24.970
I grew up. Now, I realize this name sounds a
19:25.470 --> 19:28.955
little disgusting if you've never heard of it before, but they are not really bugs
19:29.084 --> 19:32.988
in that sense. They're a type of slipper lobster from Australia
19:33.567 --> 19:36.860
and they are absolutely delicious. They're sweet, they're delicate,
19:36.924 --> 19:40.153
and honestly about as good as seafood gets really.
19:40.700 --> 19:44.350
And when they're grilled, with garlic and butter. Yeah,
19:44.494 --> 19:48.188
having that sitting outside by the water in cans, that's pretty hard
19:48.688 --> 19:52.557
to beat. And yes, I do promise they taste much better than they sound.
19:52.926 --> 19:56.540
You had me a little scared there for a minute, Jason, but that sounds absolutely
19:57.040 --> 20:00.908
delicious. Well, this was super cool. Thank you so
20:01.408 --> 20:05.084
much for joining me, Jason. Thanks so much, Amelia. If you'd like even more
20:05.149 --> 20:09.196
information about this topic, I've included a couple links
20:09.244 --> 20:14.663
below the player on this week's Fish Ryan page on eejournal.com
20:15.197 --> 20:18.818
and in the description for this week's YouTube episode as
20:19.077 --> 20:22.542
well. Hey, have you checked out EE Journal on
20:22.591 --> 20:25.932
social media yet? Well, you should.
20:26.305 --> 20:28.235
You can find us at facebook.com/eejournal.
20:30.896 --> 20:33.933
If LinkedIn is your thing— I dig it— you can follow
20:34.029 --> 20:37.661
us or me on LinkedIn, and we are also
20:38.161 --> 20:41.244
on Blue Sky Social and Mastodon as well.
20:41.999 --> 20:45.647
And we have that YouTube channel I just mentioned,
20:46.097 --> 20:50.018
youtube.com/eejournal. Folks, it is
20:50.260 --> 20:52.911
chock-full of all kinds of techie videos,
20:53.345 --> 20:57.733
including our very popular Chalk Talk webcast series
20:58.311 --> 21:02.286
hosted by me. And of course, you can subscribe to our
21:02.415 --> 21:05.280
EE Journal YouTube channel as well.
21:06.117 --> 21:11.840
Thank you everyone for tuning in. If you know of any cool
21:12.340 --> 21:15.091
new technology, or heck, you just want to chat, shoot me a line at Amelia—
21:15.156 --> 21:18.679
that's A-M-E-L-I-A— @eejournal.com,
21:20.687 --> 21:23.838
or post a comment on our forums on EE Journal.
21:24.401 --> 21:28.870
For the week of August 7th, 2026, I'm Amelia
21:29.370 --> 21:31.925
Dalton, and you've been fried.