Transcript

Introduction to Merlin Chat

I’m Dylan Manfredi. I’m a software engineer here at PMI, and I’ve been working closely with Chris Mullins on Merlin Chat. This is our chatbot extension to the Merlin analysis, and today, we’re gonna cover the white paper on it, as well as how to use it in PQ Canvas, some of its features, how it connects to the standard analysis, and some of the common use cases utility engineers have with Merlin Chat.

Getting Started with Merlin Chat

So to start, you will go to any recording that you have performed an analysis on. And this is a real recording, anonymized, so it’s real data. We’re gonna go to Chat, and your interface will go here. Now, Merlin will actually bring you back to the last chat you used.

Now in this case, if you’ve never chatted before, you will instead land here, and this is your standard prompt. It will look awfully familiar if you’ve used something like ChatGPT or Claude before, and it works in a similar way. So we can start by asking just any sort of question. A really common use case, if you don’t even know what to ask, a great thing you can ask is, “What is the worst thing happening in this recording?” This is really broad, but it’s a good first step, and if you don’t want to have to parse through all of Merlin’s reports, this is a great place to start.

Chat Interface Features

So Merlin will take a minute to think. Now while Merlin is doing that, I will go over some of the features, the UI. So on the side here, we have previous chats. You can open this up and collapse it, as well as starting a new chat.

You can rename your chat. In this case, it’s automatic, so the title is just Worst Thing in Recording, so we can go here, change that to Worst Thing in Recording 3. We can delete the chat. We’re not gonna do that, of course, but we can also export the chat, and that will give you a full transcript just in plain text, something you could attach to, say, an email.

Standards-First Approach

Now, while this is thinking, we’ll actually go back to a previous chat that has a completed conversation we had with Merlin. So in this case, we’re talking to Merlin, “What’s the worst thing happening in this recording?” Now, it starts out by referencing IEEE 1453, and I wanna highlight this, because it’s really important that Merlin is a standards-first agent. It will reference only what standards, as well as what is in your analysis, and this is incredibly important.

This keeps it grounded, and this prevents it from hallucinating or from making things up. It has the latest standards, the official IEEE standards, and it will, at every turn, try to reference those standards to the best of its ability. And this is really important. Again, this helps keep hallucinations down and makes the chatbot much more reliable when actually discussing issues in your data.

Standards Compliance and Providing Context

Now, beyond just asking, “What’s the worst thing in this recording?”, we can also talk about standards compliance. Now, in this case, it does discuss a little bit about flicker violations, and so you are in violation, in this case, of this recording of IEEE 1453.

And you can go back and forth on this and provide extra context as well, because over the course of a recording, when you start a recording, you might not know everything about the situation. But over time, during the investigation, while you’re out there in the field collecting data, you may discover additional information, like say, the weather, for example, can have a huge impact on power quality. And so it’s important then that you can provide this in the chatbot.

Merlin’s Math Capabilities

In this case, Merlin talked a little bit about flicker, and that’s interesting, but now let’s say we upsized our transformer, and the next step is that Merlin can do mathematics to answer some of your questions, as well as it’ll show its work. It’s important that it shows the math, and if we go here, we’ll see exactly that.

In this case, we’re talking about upsizing the transformer, asking if maybe that would bring us within spec. And so in this case, Merlin will go through the math, and then provide its follow-up saying this isn’t a guarantee. This isn’t going to do it, but it’s important that before it makes its judgment here, you have all of the following math. It did its work. Equally important, you can check its work. You can make sure that you can check its math and make sure, hey, does this match? Is this actually accurate?

Follow-Up Questions Beyond Standard Reports

One of the things that makes Merlin really useful is that in a standard report, there are often things that you can’t capture solely from just reports. And what we’re seeing with Merlin Chat is it helps engineers ask follow-up questions that don’t neatly fit within these reports.

You might have a really broad question, “What’s the worst thing happening in this recording?” Though from the dashboard, we can see flicker, and this is why it’s really useful, because you can dig into the details then about what’s really wrong about flickers specifically, like how off are we, and as well as what can we do to get within spec?

Follow-Up Investigations and Advice

Another use case of Merlin is actually these follow-up investigations, find out, “We’re obviously not standard compliant here. What do we do?” And Merlin can provide follow-up advice, again, always grounded within the standards, always grounded within your information. It won’t make jumps or reaches in logic. It will only discuss well within its bounds.

Learning Tool for Junior Engineers

And that’s really important because another use case of Merlin Chat is, it is helpful for junior power quality engineers. The junior power quality engineer might not be the one who necessarily makes that final call, but this can be a great learning tool, because it can take a lot of very intimidating information, a lot of data, and then compress this into an easy to learn, easy to interact with medium. And so it’s really useful if you have new engineers who need a little bit more help, and it can be a huge knowledge base for them.

Attribution Analysis

Another great use case is attribution analysis. So if you’re doing an investigation between customer and utility, Merlin can help break down ways that the customer might be potentially out of spec, what they might be doing potentially wrong, and as well as help build your case for the attribution stage.

And this is a great place too to provide extra context, because there’s always going to be little bits and pieces. Reality is often very messy. Customer might not be doing regular maintenance, or there might be something about the environment that’s just particularly caustic. And so being able to provide that information in a chat interface here is super useful to provide information that Merlin couldn’t have had during the standard analysis process.

Additional Chat Features

Of course, we can search for chats. So if you have long running chats, it’s easy to find previous information. Again, you can delete chats. If you don’t want to export the entire chat, you can actually copy the message down here, and this will preserve all the styling, specifically the LaTeX. So this is something that’s really nice. You could just drop it right into an email, send it to a coworker, as well as provide feedback.

Merlin won’t be perfect. If it makes any mistakes, you can always just down-vote that and let us know. If Merlin does something particularly good, also give it a thumbs up. We’re always looking to improve Merlin, so feedback is always welcome.

Chris Mullins on Merlin Chat

Dylan covered it pretty well. But really the key with the chat is that it’s a power quality buddy, a brainstorming buddy. You can use it to ask questions about what it sounded like already. You can tell it more context, like you can say, “This is a 25 KVA transformer,” or, “Ignore the storm that happened on the 17th, that doesn’t count. Focus on harmonics.” So you can tell it things you know about the situation that aren’t in the recording, and it’ll take that into account.

You can just ask it questions directly about the IEEE standards. That can be a useful way to use Merlin. So you can generally chat about power quality with Merlin, and it knows a lot about power quality, and it knows everything about the recording in its window there. So whatever happened in the recording, Merlin’s gonna know about it. So this is a great way to learn more about the recording and have it walk you through what’s important and what’s not.

Q&A: Pricing

Thank you, Chris. So I think that just about covers everything I wanted to talk about with Merlin Chat. And so now I’d like to open things up for just general questions.

We got a question from Casey. “I watched a video when it first came out. Do we know pricing now?” So currently, Merlin Chat comes free with your analysis. You can chat as much as you like once you’ve paid the upfront cost of the analysis, and it’s free for as long as you like. That could be subject to change, but as it stands, it is totally free.

As Dylan said, the cost to run Merlin on the file itself isn’t free. That consumes credits, and how many credits it takes depends on the complexity of the file. Generally between, say, 20 and 50 credits, which are nominally a dollar apiece. But then, like Dylan said, once you’ve run Merlin on the file, now you can chat with Merlin as much as you like.

And if you want to try Merlin, you can certainly try Merlin for free. There’s a demo that you can sign up for that you can reach from our website. So you can go create a demo account in PQ Canvas and see Merlin’s output on the files that have already been run Merlin on, and then chat with it there.

Closing

If anyone has, thinks of any questions after this chat, let us know. You can always contact us. We’re always happy to talk, as well as discuss the ways that we can make Merlin work for you. Thank you all for coming. Again, I’m Dylan Manfredi, and this was my white paper talk.