Name: Understand your Data Faster with Project Query
Uploaded: 2025-04-15T17:46:05.755Z
Duration: 30 min 8 s
Description: Understand your Data Faster with Project Query

Transcript for "Understand your Data Faster with Project Query": Alright. Hello, everyone. We're gonna go ahead and get started here. Thank you for joining us today for this Everlaw webinar. My name is Greg Marliave. I'm the VP of Product at Everlaw and then I want to under welcome you to understand your data faster with Project Query, a new AI powered workflow to investigate what's in your data much faster and more effectively. I'm joined by Simone Wasbin, the Senior Product Lead who worked on Project Query and will be showing you the product in action today. In just a few minutes I'll set some context around the opportunity presented by AI and our philosophy around building tools that leverage technologies like generative AI and retrieval augmented generation and give an overview of Everlaw AI Assistant's newest tool, Project Query. Right after that, I'll hand it over to Simone to show you how this product works. We'll also want to answer any questions you might have and we'll address some of them at the end. Before we get started though, I have a couple of housekeeping items to introduce you to our webinar platform. First note, we do want to confirm that today's session will be recorded. To re watch today's session, all you need to do is revisit the event by clicking on the same event link you used to join us live today. In a couple hours after we end the live feed, the recording will be available here to watch on demand. Now to go over a few features and resources in our webinar dashboard, I'll share a mock up of what it looks like for the audience. So, on the chat feature, the session chat is on the right hand side of your screen. Say hi and have fun chatting with other members of the audience. While you're there, don't be afraid to start off by telling us where you're joining from today, while you're dropping in your favorite GIF or emoji. If you have a question, please submit them to the Q and A tab on the right. We'll do our best to answer all of the submitted questions before the end of today's webinar. Note that this session chat is not monitored for questions, so if you want your question answered, please post it to the Q and A space. And then finally, there's the docs tab where we have several resources and links to other ways you can participate with the programs and events at Everlaw. So with that let's get started. All right. First, let's set a little bit of context about the problems that you're facing. So when it comes to discovery, many of you on this call, whether you're a legal team in a law firm, corporation, or government, face common challenges. So you have increasing and often overwhelming document volumes. This creates problems at the outset in collecting and processing that data, and that's one of the big advantages that you've probably already experienced with Everlaw. But even once you get it into the platform, then you need to act on that data to meet production obligations or extract key facts and evidence to build your story. You're often limited by available tools and workflows, like traditional keyword search methods that can require very complex and advanced searching logic. They can miss key concepts when there are synonyms or misspellings that affect the data and what's returned. And the results may be too broad and have a lot of noise, which can create added, pressure on your workflow. And then you might have time and and cost considerations as well to get to that first insight. You might need to make sense of hundreds or thousands or even millions of documents, and and you're working against tight deadlines and budgets as well. And this is where we think that, AI and Everlaw can really help you in this. And so starting with just a little bit about our philosophy for building, like, one, we try and always build for the long term. We want it to be adding additional functionality that continues to bring more and more value for you. So we wanna take the time to ask the right questions in order to create, you know, high quality litigation and investigation management tools that really understand your workflows and what you're trying to accomplish. Our goal with any new tool, including within Everlaw AI, is to apply technology in a way that drives real and practical value for you, our customers. And in this case, with query, we're talking about creating high quality verifiable answers to important questions, at any stage, but especially early in an investigation or litigation. Okay. And so, this is, exciting technology for retrieval augmented generations and LLMs and generated a overall, and we have a number of AI tools and so Project Query is gonna be the newest addition to that which we recently released into beta, but it goes alongside a number of other tools that we have both in our core product offering with translations, clustering, and predictive coding And then also within the Everlaw AI assistant that's already available, to you in live and general availability with things like coding suggestions, writing assistant, and review assistant extractions. This is a suite of tools that helps you starting from review, surfacing evidence that may not have been uncovered, or significantly reducing manual review costs through to case building, helping you to use that evidence to tell more powerful stories. Let's dive in to a little bit more detail on this latest innovation code named Project Quarry, which again is now available in data. So the specific value that we're focusing on here, is finding the key evidence and critical insights in in massive documents that has always been very challenging for customers. It's time consuming that leads to significant costs, and it also takes a lot of time, which can delay the discovery of key evidence until after after important moments in the case where it could impact your outcomes. And so what we have is a tool that lets you ask natural language questions directly on entire datasets, to get answers supported with rabbit with references along with relevant facts and key documents much faster than you would be able to accomplish sifting through that data manually. And so what that workflow looks like is you'll be able to easily ask these questions in natural language across again an entire corpus of documents, with open ended questions the same way that you would a colleague or a really smart intern like we like to talk about for our AI assistant overall. And so you won't need complex searches or deep techno technical expertise. You'll get precise citation backed answers in under a minute, allowing you to verify those results, and you'll also receive a range of relevant documents and facts alongside that answer to further expand your understanding and guide additional exploration. So this allows you to investigate a corpus of documents confidently with answers grounded in the four corners of the document set with easily verified cited sources and securely with no client data stored or retained by the LLM. To walk you through in a little bit more detail what that looks like from from start to finish, when you're you're gonna ask your question and we're then gonna use this retrieval augmented generation workflow to start by understanding and interpreting what your question is using an, embedding model and LLMs, and then we're gonna send it through to run that that query through embedded storage, to find relevant excerpts within that really broad dataset that likely contain the important facts for the question that we're gonna need to answer it, and then we'll send those through to an LLM for that final pass to then take those relevant excerpts, consume that, and generate the final answer with citations back to the documents where those excerpts came from. And this is a tool, like we mentioned, that's already available in beta. So Simone and our team have already been working with early access users for real practical value in the application of of Quirion live matters today, and we've gotten really great feedback about that. So they've experienced how fast the tool is, how it's able to, you know, create practical value at extracting insights early on in a case, and doing so, you know, at game changing second speed is second speed, excuse me, is a really big differentiator especially early on in the case, but that can come up at any moment throughout when new things, new things are discovered through the course of of your litigation or investigation. Alright. And with that, I'm gonna turn it over to Simone to show you Project Quarry in action. Awesome. Great. Thanks so much, Greg. As Greg mentioned, I'm Simone. I'm a senior product lead here at Everlaw, and I'm gonna share my screen to do a quick demo. Great. So I'm in the results table in an Everlaw database, documents contained, produced in a massive opioid litigation. So we see here there's 1.3, two five million documents. And as Greg mentioned, you know, a challenge that essentially every investigation or litigation team faces is, you know, finding those needles in the haystack, finding that key information, quickly and efficiently. And Project Query is a a tool that again uses retrieval augmented generation, a very refined rag approach, to transform the way that legal teams can interact with their document sets. So we've designed this tool for flexibility. There's kind of two ways you can enter the tool. You can enter, using the universal navigation which defaults to asking questions across your entire 1,300,000 documents in this case or you can enter the tool and ask questions across any results table set by clicking the button on the top of the results table. If you have a binder or a specific keyword search or date range search in mind, it's really customizable to drill down on just any subset across the 1,320,000 documents. I'm actually going to get started asking questions across the entire project database here. And so in this opioid litigation, there, you know, were some claims of allegations against opioid manufacturers for contributing to the crisis through their questionable sales tactics and in an inadequate or monitoring. So let's say, pretend I'm in kind of an early case exploration mode. I'm gonna come here and enter my first query. Asking a question like, how were pharmaceutical sales representatives incentivized to increase opioid prescriptions? Just trying to get a general sense of some of the key facts that might contribute later on, see what documents are accessible related to the sales representatives incentives. I'm going to go ahead and click Send here. Now we see the responses starting to generate. I'm going to explain what's going on behind the scenes, whatever law is doing to get to this response. It's not instantaneous necessarily like an immediate chatbot response, but it is pretty quick, usually within seconds under a minute. The reason for that is that Everlaw is taking three steps behind the scenes. First, Everlaw is identifying potentially relevant documents. So these are documents potentially relevant to the question that you're asking. Second, from those documents, Everlaw extracts key pieces of information or key facts from those documents and analyzes them for relevance to your question. And then finally, from the most relevant pieces of information, Averola generates this final output that we just saw pop up on the screen here. So we see a pretty lengthy paragraph detailing the incentives that sales representatives were given. So they're given like performance based compensation plans. Many provided modest base salaries with bonuses or commissions. And then we see at the end of this kind of first sentence here, these blue hyperlinked references. And so hovering over the reference shows the document, controller Bates number in addition to the extracted back from that document, as well as the option to open and view the document. We see that in this first example, there's more than five. There was six total. We can see the additional references by clicking into it. This is super important to call out because information used to generate the responses coming from a third party sites, from the Internet more broadly, it's all coming from the document set across which you are asking. Then I'll point out, in addition to the results we're seeing here, we see the resources expandable tab down here. There are three components to resources and these correspond to the three steps that Everlaw takes to get the generated output. I was mentioning those three steps earlier. First, Everlaw is identifying these possibly relevant documents. Out of the 1,300,000 documents, Everlaw identified 140 documents as being potentially relevant. From these documents, I can open them in the results table here and perform any other actions I would in Everlaw. I could batch them out for review. I could apply other Everlaw AI like performing or getting extractions or coding suggestions on them or summarizing kind of the key documents here as well. And then from those 140 documents which we give you flexibility to do with whatever you want with them, There are 66 facts generated from them. Thinking about it like a funnel, we're starting broad with the most potentially relevant documents and then pulling out the key pieces of information, the most important information and the most important documents to answer your question. And so in this fax table, we see the reference, which is the first column and this corresponds to the reference number if it was used to generate the response. So in this first row here, we see the reference number one. This is referring to the fact corresponding to this one here. And then the second row, there is no reference number. So this was another highly relevant piece of information that just didn't make it into the response. It wasn't determined to be the most relevant. And then we also display the relevance score here. So this is a number one to five. One being somewhat but not very relevant and five being highly relevant. And so this is something that you could drill down on if you wanted to explore the additional information that Everlaw thought to be somewhat relevant but not relevant enough to make it into the final response. Then the generated fact, which again is the piece of information that Everlaw has extracted from the document that's relevant to your question here. The actions you can take are to open the document, you can filter this table to drill down on what you care about the most, search through its contents. Then the facts used in response. From the 66, all generated facts, Everlaw identified 19 to be the most highly relevant ones to your question which were used to generate the output above. This table shows the same information that the all generated facts table shows. It's just a subset of the all generated facts table showing just the facts that were used in the response above. Rather than having to hover over each one, all the information that was used in the response is displayed in an easily skimmable table which is highly customizable to filter and drill down on what you care about the most here. So I'm gonna queue up some other questions, to kind of drill down and show some some more examples. So let's say, later on maybe perhaps, further down the line in this litigation, I kinda wanna find some some more specific examples. Maybe ask a question along the lines of supporting some some claim my team might be trying to make, you know, in a in a trial or just piecing the story together. What evidence do I have to support that there were oxycodone orders that exceeded the DEA guidelines in terms of quantity? I'm going to go ahead and submit that question as well as queue up another here. Another question I want to show to demonstrate that Everlaw, if the question is not relevant to the underlying documents, if there's no information related to the question that you're asking, Everlaw will indicate that. I'm going to show this through this example of the question, why do cats have stripes and send. You'll see here for this question, that there are, you know, this is an opioid litigation matters, documents produced related to an opioid case. There's nothing really relevant to cats having stripes and so we'll see that in the response. And this is really important to call out, and kind of grounds the the Everlaw AI principles of competence and control. We want to make you confident that the results that you are getting, are verifiable and you're able to verify that the information that's being used is coming from the underlying document sources. You saw that that response generated more quickly than the one that had a lot more information that we were doing more analysis to synthesize. We knew pretty quickly from the start once we tried to identify those relevant documents, that nothing was really critical to getting to the question you had here. Going down to this drill down question, where are there oxycodone orders exceeding DEA guidelines? So we see that there were multiple excerpts indicating that several oxycodone orders far exceeded guidelines. And then we're seeing additional details related to this. Right? Well beyond the DEA thresholds, for instance, some order of 70,000, units per month, controlled, over the controlled substance limit by tens of thousands of dosage. So these examples are are super interesting to kind of maybe drill down on you. We see the there are some some more examples here, giving us even more specific kind of numbers if we wanted to kind of drill down and find, you know, some documents that, you know, again might be supporting evidence to a claim my team is making. So for example and I'm showing how you can, open a document here. This kind of the relevant fact from this document, was related to the fact that the oxycodone guidelines were around sixty nine thousand doses in this case. And this pharmaceutical company had, as we can see here, dosages far exceeding these guidelines. And this was, you know, on page 11 of a longer spreadsheet, something that wouldn't have been found with traditional keyword searches most likely or would have taken a lot more time than, you know, there's probably thousands of documents that hit the the search term DEA guidelines and dosages. Thinking about different ways you might go about searching that. So that's just one kind of really powerful example to show that, you know, beyond the the keyword, kind of searching that how how powerful this technology is here. Right? And so, yeah, I guess a couple other things I wanna walk through with with kind of how we've designed this in the interface, is that we've designed for it to be flexible and used throughout the matter life cycle. I gave a couple examples of where you might be thinking about using it a little bit more exploratory versus drilling down on some key facts. So we retain all questions and allow you to search through past questions, resurface kind of those critical insights that you discovered earlier in your case, as well as make the tool collaborative by being able to search through other questions that folks on your team had asked if anyone else had asked a question here as well. As Greg mentioned, we're just so thrilled that Project Query is released in a closed beta and being used to help actual legal teams understand their data faster and uncover insight in a totally new transformative way. And we've been really pleased with the positive feedback our clients have, you know, said so far for the folks using the tool. We've heard some, some of our clients use the word, game changing, as well as solid gold. Just having a totally new transformative way to surface insights and interact with their datasets. So I think that's that's all I have to really share for this demo. And I guess I can go on to, a few questions, that have been asked by the audience here if we have a a few minutes. Just checking the time. Yes. One question I see coming in is related to what types of questions the tool is better at answering. That's a really good question here. We see this project query is one tool in the Everlaw AI toolkit, not necessarily, you know, the the the one magical piece that addresses all your problems and and just solves your case for you instantaneously. Right? So project query was designed for precision. It was designed to help you find answers to either like open ended or specific questions by surfacing those most relevant pieces of information. Project query is not intended to replace traditional keyword search or aggregate information. Types of questions that you might use the words in like, please return me all documents related to x topic or containing x keyword. Typically, those will be more appropriate for your traditional keyword search. I'm not saying project query wouldn't be helpful to surface the most relevant and some give you a good place to jump in and start from. But again, it won't be entirely comprehensive in the results it is surfacing for you. Then I see another related question here about thinking about Project Query versus using other Everlaw AI tools such as coding suggestions. I think one thing again to emphasize is the scope of the results that you're getting here. With coding suggestions, you're getting that suggestion or even for custom extractions, that extraction for all of the documents that you are running the task on in batch. That's going to be a very comprehensive results or categorization of the document that hits on each and every document. For examples where that's what you're intending to do, you're looking to get to that level of granularity, for that comprehensiveness, coding suggestions is great there as well. Then thinking a little bit more downstream too about writing assistant. I've had some clients ask, you know, is project query intended to to replace the writing assistant? And the answer is is no to that. We see, the writing assistant, you know, being really critical for helping you draft and jump start drafting workflows to, you know, for for more complex and more detailed analysis. Right? The information we're getting here isn't like a full comprehensive memo on the oxycodone orders or the the sales representatives. Though it does give you a really good starting point to find information to feed into the writing assistant. Thinking about maybe as I'm preparing for a deposition, identifying the key documents related to the deponent we're deposing and the issues that we want to ask them about, finding those documents, surfacing them through project query, and then using the writing assistant from those documents to draft deposition questions or to anticipate deposition questions that a deponent might be asked during a deposition. So that's, yeah, a really great question. We're really excited to see how these kind of tools kind of like fit in fit in together and enhance your their workflows, in in in unison and, in harmony together. And then I see another question, related to about risk for hallucination in the answers. So that's a really good question, because, you know, with any generated AI tool, there's almost always a non zero risk of hallucination. But one of the really powerful things about the rag technology is that it dramatically reduces and minimizes the risk of hallucination, by grounding the answers in the underlying documents here. So we can see that each piece of information, each sentence is backed by these citations. This really emphasizes, again, the Everlaw principles of competence and control, giving you confidence that the responses are not being hallucinated and giving you the ability to go in and verify the reference information. And so while the the risk of hallucination is not necessarily non zero in that way, it's it's very minimal minimalized and contained, by by the underlying technology we use. And we've really refined the approach and to minimize this and to give the answers fully referencing the underlying documents across which you are asking. Okay. I see a lot more questions. I think I'm running up on time here but the question that I'll turn to next is about when will Project Query be more widely available. As we mentioned, Project Query is currently in a limited closed beta and we don't yet have a date for our general availability launch. But this is a great time to pass it over to Greg to close us out and share more details on how you can learn more about Project Query. Thank you. Alright. Thank you, Simone, for walking us through the product and answering some audience questions. For the questions we didn't get to today, we will follow-up with you directly after the webinar. So I hope you've taken away with you a few things from the webinar that will help you, start to envision how tools like Project Query can help benefit, your firm or organization. And before we wrap, I also want to invite you to join Everlaw's community roundtable. This will be an interactive session exclusive to Everlaw community members to have a deeper discussion with our product team about ProjectQuery next week, Tuesday, April 22, at 9AM Pacific time twelve eastern. And also to join the waitlist. So if you are interested in Project Query beta, you can add yourself to the waitlist @everlaw.comslashbeta. We'll have links in the doc stab which will allow you to join the community and beta wait list as well. So thank you all very much for joining us and hope you have a great day.