Video: Unleash the power of Nutrient AI Assistant | Duration: 1448s | Summary: Unleash the power of Nutrient AI Assistant

Transcript for "Unleash the power of Nutrient AI Assistant": Hey, everyone. Thanks for joining us today for our final webinar of 2024, where we'll be giving an exciting demonstration of our newest offering, AI assistant. I'm Greg Ibs with the product marketing team here at Nutrien. And joining me today is our VP of product, Milos Jekic. Milos, always a pleasure to present with you. Thank you so much for not butchering my last name, Greg. You know I rehearsed that many times. For a southerner, that just doesn't roll off the tongue. Well done. Alright. So before we jump into our presentation, I do need to cover a few housekeeping items. Today's event will last approximately 20 minutes or so. If you have any questions during the webinar, just find the q and a functionality on the right side of your screen. You'll also notice that you have a place for chats as well as the q and a. Please use the q and a functionality. You can also use this feature to let us know if you're having any issues. Alright. And finally, yes. The webinar is a popular question. The webinar is being recorded and will be available on demand on our site. Alright. On to the agenda. First, many of you attending might have missed our recent announcement. It was a pretty big one. It was around a company rebrand. So, to start things off, I'm gonna give a 2 minute overview of the new company and why we rebranded. Then we'll jump into why you're here today, and that's to learn about AI assistant. And then after a quick overview, I'll hand it over to Milos where he'll run through a demo of some of our common use cases. And at the end, we have saved a little bit of time for a few questions. Alright. So to kick things off, you're likely aware that we made a major announcement recently with our company rebrand. Whether you knew us before as PS, PDF Kit, Mahembe, Orpolis, Aquaphorcer, even Integrify. We are now Nutrien. Same company with the same technology, just rebranded under a single banner with a single mission. And that is to evolve how humans experience and interact with documents. With the rebrand, we now have an unrivaled suite of document solutions all in one roof. Whether it's the industry leading SDKs or our M365 low code solutions or our workflow automation platform. We provide tools that serve as the building blocks to create seamless, secure, and scalable document ecosystems, Tools for developers, tools for product leaders, and tools for IT leaders and citizen developers. Nutrien was born of PSPDFKit, backed by Insight Partners and accelerated by the leading innovations of Orpolis, Mahimbi, Aquaforest, and Integrify. We take pride in simplifying the complexity that you face with speed, savings, and security to become your solution for today's document challenges, but also, more importantly, your partner for tomorrow's document innovations. We are Nutrien, and we're thrilled to be building the future of document innovation alongside of you. Alright. Which brings us to today's webinar. You know, as a product marketer, I get to see I get that front row seat to all the new features and offerings we roll out as a company. And I can safely say that none have generated the excitement that we're already seeing with AI assistant and the potential that this brings to the market. If you joined us a couple weeks ago for our q four product update webinar, you heard me talk about how much the world has changed since being introduced to OpenAI's Chat GPT, which coincidentally just hit the 2 year mark last week. Now you see AI use cases popping up everywhere. And that's especially true in the document space where everyday users have grown to expect native capabilities that allow them to interact with documents in front of them. And that's why we built AI Assistant. However, as you'll see in a few minutes, this powerful new tool goes well beyond simple document interaction. Imagine giving your users the ability to collate information from across your document and generate a comprehensive summary in seconds. Or letting them talk directly to a document in natural language and get an instant precise answer without sifting through pages of text. With AI assistant, there's no need to comb through dozens or hundreds of pages to find the data you need. The information comes to you. But again, that's just scratching the surface with the type of capabilities that you can now provide. For instance, imagine being able to redact all of the sensitive information from a document just by saying, redact all private information for me. AI assistant contextually understands what private information means and will remove all sensitive content instantly. But that's not all. AI assistant also allows users to search, reword, or reformat text instantly. It can also translate entire documents or portions of the document into multiple languages on the fly. Need this paragraph translated to French? No problem. Done. The amount of time and effort that AI assistant will save users is extraordinary. But don't just take my word for it. Let's see it in action. I really can't wait to show you everything. Prepped a couple of documents for you all, so could be fun. Let's dive in. What you're looking at here is Avolon. Aveline is Nutrien's end user cloud based document viewer which has our AI assistant built in. When I open up a document and this one is actually Japanese tail, I can find the AI assistant and engage it right here from the from the bottom right corner by clicking on this icon. Just to be clear I do not read nor speak Japanese so I have no idea what this is about. When I open up the assistant it automatically sends some suggestions on what I could do with the document. And since I have no idea what this is about, I'm gonna select the first one which is to summarize the document. Suddenly, I see what the story is about. There's a kid. He grows up very small, 1 inch tall, but ends up slaying the demon and obviously getting the girl. Kind of reminiscent of a story that I heard when I when I was a kid. So I'm pretty sure you've already noticed, but the the the response is streamed which is reminiscent of chat gpt. It is the exact same experience when you're chatting with an LLM with one of the popular tools. But again, this this is happening inside a viewer that can be inside of your software so your users don't have to actually download the document. Go to chat gpt and ask a question like this. Okay, Japanese, this was a summary. I'm interested, how about if I do this? I get a context menu. Look, it's the same icon that was used to engage with the chat and I can grab a selection and do something with that selection. Let's say I wanna translate to English. Now I get an actual full translation of this text to English. As we know, l m's are really, really good at translating. You know what else LLMs are very good at? Just giving some responses about different things. So let me see if I can do this. What is the best, mojito recipe? Tell me. Ah, it doesn't wanna do it. So this is very important actually. We implement a lot of guardrails so that users can't really abuse the AI assistant for their own needs. You know, they're not gonna end up spending a whole lot of open AI credits. Or if you're using like a self hosted LLM, they will simply not abuse that for their own purposes. What's really good is, you know, the answers, if if somebody's doing q and a on a document, will happen within the context of the document or an expanded context that you can provide additionally, of course. I wanted to I really love this story. Maybe I could put it on Instagram. So let me try this. Make an Instagram real script based of based on this tale just to see if he's gonna do it. Scenes, narration, voice over. Amazing. Amazing. You know what? I really loved this tale. So let me grab something from the Gutenberg project, which is a collection of Japanese tales. I can already see a useful question here. What themes are prevalent in Japanese fairy tales? By the way, this is quite a big document, so let's see how it works with a big document. Pretty good. Pretty fast. Oh, okay. Now I know that there's bravery, ingenuity, moral lessons, consequences of actions, greed and perseverance, transformation, and discovery. Pretty cool. I can also kinda click to see where this information is being grabbed from. You know fairy tales from my childhood often use animals that sort of, you know, convey a sort of a personality. They personify different things. I'm wondering if it's it's the same with Japanese fairy tales. Let me add this. What animals are mentioned in this document? Cool. There's a bunch of them. Centipedes, Parrot, Badger, Boar. Cool. What are some of the main characters of these tales I can't spell? Cool. Some of the main characters are these and that. Momotaro. I like that name. Momotaro. Let's see. Find where Momotaro is mentioned. See if we can find those tails. Amazing. Look, it even tells us what they're known about. So I can just go to these tails, find them, and and deal deal with. I guess where I grew up with Disney, you know characters like Bambi, Japanese kids grew up with Momotaro. But let's see how we can use the assistant with something more serious. Let's say a reseller agreement. So recently, we entertained becoming a reseller partner of this certificate provider. And, you know, we got a whole lot of legal texts to review, etcetera, etcetera. But, you know, we had these very specific questions that we needed to answer. Right? So cool. What I've used RTech, which was at the time in development, and the first thing I did obviously was and I always do that now, I just summarize the document. So this document gives me all this information. Right? Amongst the questions was, how much commission are we gonna actually pay if we become partners? The important thing is now that I've learned the context of what this document is about, I know it's between a DSP and a project partner, I can tell it that I am the project partner or we are are the project partners. How much commission do we have to pay to the DSP? And I can instantly get the answer without scrolling through a potentially very long legal document, of course, and it will tell me exactly what this is. It doesn't specify. Instead, it states that it depends on the product line, current commission levels of detail in appendix a, which is not provided in the document. Great. So I I can find that information in the appendix. Boom. I know what to do. This is where I can I can pinpoint, this text is actually leading me to appendix a? The other thing that I was looking for is what exactly what services, would we pay commission for. Right? What exactly are we paying commission for? And this is how I got an instant answer here because I do not have to read the damn document, right? Certificate products and trust services, I can obviously click to find to find where this information is, navigate to that page, etcetera, etcetera, to find that information. Okay. So the last but not least was, the questions that I had was, can we sell these services under our brand name, and the information that it got me very quickly was, yeah, you can, the agreement allows for joint sales marketing, close the promotion, etcetera, etcetera. Cool. There are subterms that are outlining the agreement, etcetera, etcetera. Now when you're going into a document like this, and you know what information you want out of it, you you can just ask. Right? Just imagine yourself expanding the context of this document with metadata, like where is this documented in its life cycle, what is the user's intent with this document when they open it up as a part of a workflow in your application. It's just you can help the user get instant value from this thing very, very quickly. And this is what we want you to be able to do with the AI assistant. Cool. So a plug and play chat about your document functionality for any web based document viewer, right, any web based document workflow that includes our viewer, and, soon enough, honestly, any mobile workflow. Yeah. Your users no longer have to download the document to upload it to chatgpt or Claude to chat about the document. Sometimes you don't even want them to be able to download anything. Sometimes you want to protect the document completely, which is what you do with our document engine, which this assistant works very well with. But, you know, we've gone way beyond what child gpt and the likes can do with a document. So I will now show you one thing that they can't. I prepared this complaint, and I'm pretty sure we've all seen something like this or even sent one of these to someone. Right? So Alex Johnson here is complaining about the TV they purchased something something. Obviously, I don't even have to read because I just engaged the assistant, summarized the document, and it will tell me what the problem is. Right? I figured out it was a sample complaint letter. Yeah. Because I used the sample. Cool. The point is this prompted me. You know? They they say they will contact the office for fake trading if I don't do something about this. Right? So let's say I wanna now send this to my technical team that claims that this TV is indestructible, it never breaks, customers never have a problem with it, blah blah blah. Cool. But I can't really send them a document like this without removing information that they can use to contact Alex Johnson. Right? They shouldn't be able to infer who the customer is. Right? So I need to redact some information. So how would you do this today? No trend technology supports reduction across the board, web, mobile, etcetera, etcetera. And this is how reduction looks like. I open up the document in a reduction, view in a reduction session. I click the the reduction tool basically and I pinpoint all the things that I want redacted. Since this is a PDF, it can be very tedious. I'm already bored. What else do I have to do? He's disappointed. Blah. Let's say I picked it all up. Maybe I want this. Oh, maybe I don't want it to know when whenever that doesn't matter. Then when I'm when I'm finished, I apply reductions. Great. Now imagine if this was a 15 page document or a 100 page document, right, and imagine you're doing those reductions, right, I mean you can search find find and redact, as well, but it's still very manual. We have an assistant. So I can just tell the assistant what I want. I want to anonymize this complaint sent to my tech team. They should not be able to know who the complainer is nor contact them in any way. Redact all information they can use to do that. Let's see what happens. The beauty of this process is the AI assistant stages these redactions and then gives me the ability as a human to quickly glance over what it got staged, maybe make some last minute changes like I want them to keep the product name, for example, I really don't want to redact that because I want them to know it's about the smart TV. Right? But everything else, pretty cool. I can just do this in a matter of seconds rather than spending minutes on it. Or if it's a 100 page document, you know, take a wild guess. And then I can, either discard changes and refine my prompt because this is an LLM, which is the beauty of LLMs. I can just apply reductions, and I'm done. If I am looking at these kinds of documents every day and they have very similar structures and whatnot, I can refine a prompt that I can keep using all the time. And what we want you to be able to do is to use these things headlessly as well, which means without the viewer. So you can refine and train a prompt and then maybe use it at scale. Alright. Thanks, Milos. Every time I see it in action, I'm just amazed. Really well done. Now before we wrap up, I wanted to remind everyone that AI assistant is seamlessly built into our industry leading document viewer and easily works with OpenAI or other LLMs using an API key. That includes privately deployed models that you might be using. I, I know that's important for a lot of our security conscious companies out there. So we are checking that box. And while we're currently only offering this on web, expect to see AI assistant coming to mobile in the near future. Alright. Well, that brings us to the end of today's presentation. Milos, what can I say? Amazing job yet again. And to our attendees, if you have questions about what we just covered, go ahead and drop them into the q and a now. Or if you'd like to, talk about your specific needs with one of our team members after the webinar, we'd love to hear from you, whether that's reaching out to our customer success team or sales team. We also invite you to Nutrient. Io to learn more about AI assistant and all the other solutions we provide. You'll find a ton, I mean, a ton of helpful resources like our blog, our product pages, our component pages, guides, demos. You can sign up for a free trial on our solutions to test things out for yourself as well. Alright, Milosh. Let's see how many questions we can get to. It looks like the first one is, is this using OpenAI? Yes. What you've seen in the demo is using OpenAI, provisioned via OpenAI API. It can also be provisioned via via Azure OpenAI, and we also support Lima 3.1, and we support Mistral large. Just wanna add deploying deploying and managing models is is is hard. And this is why we also support AWS Red Rock. It's a it's a model provider with access to LAMA 3.1 various other models. I guess, thanks, AWS. But also our customers can use an ALN that they host themselves. Excellent. Next one. So do we just use our own OpenAI API key? Yeah. So if you want to use the assistant with OpenAI, you will need to provide the API key. I guess if you're worried about cost, how you're gonna manage that and all that, you can configure the back end of the assistant, to control and do things like cap usage for different for different users of your software. Okay. Next one, we're using your web SDK. How can we integrate the assistant? Okay. So at the moment, you'd also need to be using our document engine, which is our service that provides server backed functionality for the web SDK and things like maintenance and collaboration for web and other SDKs, and things like headless document processing, for example. And then you would basically deploy the assistant as a sort of a sidecar to the document engine or the web SDK. You configure it to work with OpenAI, for example, with your keyword, another LLM, for example, one that you're hosting. And then basically, it's a matter of just toggling it on, in the UI. Both the assistant and the document engine are basically Docker containers, so there's many different ways you can deploy them in your infra. And right now, we're working on removing the document engine as a as a requirement here so that our customers who use our web SDK in standalone mode without the backings of of that server component can also deploy it as a sort of a standalone sidecar to the web SDK. And this is also how this is going to work, with with our mobile SDK. It's basically you would just deploy that Docker container and point the mobile SDKs to or the web SDK to its address, essentially. Great. Excellent. We have time for one more. Let's see. You said we can use our own privately deployed LLMs. Which ones? Well, as long as the ALM supports the same API spec, as OpenAI, which most of them do, you can use them with the AI assistant. However, I do wanna point out we've seen the best results with GPTs and llama, 7 o b. It's all it's all well documented in the guides, by the way. Alright. Well, that's all the time we have for today. If we didn't get to your questions, no worries. We'll follow-up directly after the webinar. On behalf of Milos and the entire team at Nutrien, thank you so much for joining us. Enjoy the rest of your day. Thanks.