Everlaw Launches Everlaw^AI Assistant Beta Program

In an office in Oakland, California, Mondee Lu has been working for months on a bit of AI esoterica. Her goal: to harness a generative AI algorithm built on nearly endless amounts of original communication and refined by rigorous human review, capable of hitherto unheard-of feats. Generative AI can draft Spenserian sonnets on command. It can prove out the most difficult mathematical theorems. And it can write human-like verse with incredible fluency. 

But can it be convincing to lawyers? And more importantly, reliable for lawyers? 

That, it turns out, is an achievement easier said than done. 

To make sure that her AI capabilities can serve today’s legal professionals, Mondee agonizes over obscure nuances to AI prompting, which elements of a request to emphasize, which parts of a document corpus to focus on, and dozens of other variables. 

These small details can have incredible impacts. A tweak to the instructions here becomes dozens of iterations across dozens of applications. A shift in implementation there means weeks of refinement, retesting, reimplementation to ensure that outputs are well crafted and precise.

A lawyer by training and a product manager by trade, Mondee has to understand – and make the AI understand – the intricate nuances of emphasis, tone, and legal application. 

And she has to make it useful, trustworthy, and almost instantly verifiable.

Mondee is just one of the dozens on the Everlaw product, engineering, and design teams who have spent countless hours refining the application of groundbreaking generative AI to the litigation and investigations process. Among her colleagues are Kevin Kraftsmith, a former attorney with the San Francisco District Attorney’s Office turned product manager, who applies the exactitude and meticulousness of a prosecutor to finding the most natural, beneficial integrations of AI; David Hirschberg and his team of engineers, who bring a scientific rigor to testing out each implementation of artificial intelligence; and Mischa Noll, who has coordinated security clearances, privacy review, and legal negotiations to ensure that the data of any user of Everlaw’s generative AI stays protected. 

They gather in conference rooms on the 16th floor of 2101 Webster Street in Uptown Oakland. They test new theories, coordinate across platforms, document, iterate, experiment. Consumed by an urgency to bring the benefits of generative AI to their audience in a way that is responsible and trustworthy, they work too late, they think too much. 

The result is magic. Days, weeks, weekends, months spent to ensure dependable, trustworthy, generative AI capabilities that can save days, weeks, weekends, months, and, in time, years in legal work. 

All so today’s lawyers, investigators, and litigation support professionals can unlock insights that never could have been seen before. Can uncover their most relevant information faster than ever. Can capture the connections between parties, patterns, and concepts. And make sense of it all with cutting-edge, generative AI built by AI experts that they can trust.

If AI promises to transform the world of legal work, these are the humans who make it possible, who ensure that users remain in control of their data and process, who create outputs legal professionals can use with confidence, who ensure that transparency, privacy, and security are at the foundation of our AI tools, and bring a thoughtful, intentional, incredibly purpose-driven approach to generative AI tools that the legal professionals can rely on. 

Bringing the Power of Generative AI to Litigation and Investigations

The end product is Everlaw^AI Assistant, a revolutionary application of generative AI to the legal sphere, announced today by Everlaw.

Everlaw^AI Assistant joins a suite of established AI tools now united under the Everlaw^AI moniker, including Everlaw^AI Predictive Coding and Everlaw^AI Clustering. 

With Everlaw^AI Assistant, legal professionals can leverage the power of generative AI, built on powerful large language models, to provide insights at the document and narrative level in ways that will transform the litigation and investigations process.

Everlaw^AI Assistant keeps human intelligence at the forefront, empowering legal teams to exercise their expertise and judgment. To allow legal professionals to quickly check for correctness, AI output is supported with direct links to sources. When the human mind captures nuances AI may miss, users can quickly edit the tool’s insights to make additions, expand the reasoning, or provide additional detail. 

Transforming Ediscovery from the Document Up

Everlaw^AI Assistant is integrated into two key workflows, the core review window where doc-by-doc analysis, tagging, and redaction occurs, and Storybuilder, our post-review drafting and narrative-building platform. 

“We know the legal profession holds its members to the highest standards,” says Everlaw engineer Matthew Kaplan. “As someone with a background in machine learning and AI, I want to make sure our technology meets those standards. I’ve been beyond impressed by how our team of engineers at Everlaw have worked so quickly and meticulously to integrate the generative AI model into our platform.” 

Document Summarization

At the document level, Everlaw^AI Assistant helps legal teams move quickly through first-pass document review to the merits. Take, for example, a reviewer who needs to understand and code a 200-page document as part of their review. Read start to finish, that would take the average person over 200 minutes, though only a fraction of the document may be relevant to the matter at hand. 

Everlaw^AI Assistant, built on transformative large language model technology, can help surface the most important information from a large document quickly. Users can leverage Everlaw^AI review assistance to create quick summaries of large documents. And those summaries can be doled out into multiple items – a summary for every five pages, for example – so that they remain informative without becoming reductive. And they are created all within seconds. 

“We knew we needed to provide consistently valuable responses in an environment that cannot be tested with private data,” David Hirschberg explains. For his team of dozens of engineers, that meant rounds and rounds of testing and validation, to ensure that “the right data was used and the outcomes were relevant and useful,” he says.

Entity Extraction and Sentiment Analysis

Of course, knowing what a document contains is only part of the challenge. Reviewers often also need to know not just what was said, but who was aware of it. 

With Everlaw^AI Assistant’s entity extraction feature, users can quickly identify the people and organizations referenced within a document, then jump straight to where those references occur, then set the entity as a hit highlight to easily move through all appearances of that person or organization within the document. 

“To ensure that the information produced is the most helpful to our users, we had to think hard not just about what the AI generates, but how the response is structured,” Everlaw engineer Yuchen Liu says. “We experimented with different prompts and improved our parser in order to accommodate some edge cases.”

Apply sentiment analysis, and users can then see an overview of the sentiments expressed in a document, both negative and positive, and navigate to representative snippets to dig into the details further. 

Open-Ended Queries

Everlaw^AI Assistant can also respond to direct questions about a document. Users may simply ask Everlaw^AI Assistant a question about the document before them, and they’ll receive an answer on demand. As with so many of Everlaw’s AI capabilities, verification is built into the core of the experience. Users can simply click into the pages where answers were generated from to dive deeper into a document or confirm an answer's accuracy. 

Each of these capabilities are carefully designed with strong guardrails, to include both easy access to the underlying information so verification can be fast and simple and to reduce the likelihood of potential inaccuracies. 

“Building on top of powerful foundational language models is quite different from other types of features because it's so comparatively fast and easy to get to an early stage feature,” Mondee explains. 

“But that last stretch to useful, stable versions of these features that we’d feel good about putting in front of customers is actually a lot of work, with a lot of potential pitfalls, requiring careful thought and design.”  

Bringing It All Together with AI and Storybuilder

Of course, document review is just one step in the long process of bringing the truth of a matter to light. Moving from a collection of potentially hot documents to a cohesive theory of a case is some of the most difficult work litigators face. Everlaw’s Storybuilder provides the platform where this challenging, creative work happens. Storybuilder allows legal teams to craft case narratives throughout the review process, managing litigation end to end with an integrated suite of post-review tools. 

Everlaw^AI Assistant brings even more power to this platform. Key documents collected through the ediscovery process can be instantly summarized by artificial intelligence, to be organized in case timelines and narratives. 

Those documents can then make up the substance of AI-powered writing assistance.

The goal is simple. As Mondee puts it, “We want to make it easier to get to the heart of the matter quickly and be able to articulate it more effectively.”

Writing Assistance

For legal professionals looking to create a factual narrative based on their key documents, whether a statement of facts, deposition summaries, or memos, Everlaw^AI Assistant can quickly turn a blank page into a strong first draft. 

Everlaw^AI Assistant’s drafting capabilities can quickly create a cohesive factual narrative based on the evidence your team has gathered. For example, legal professionals can prompt the tool to construct a statement of facts based on the documents identified within Storybuilder. Let the AI run automatically, and it will produce a straightforward overview of the key facts. Documents are automatically included as in-text citations linked straight to the source of the information.

The Everlaw team deliberated over not just the substance of these outputs, but the style and tone that they take. “The role we ask the model to take needs to be appropriate to the task,” Kevin Kraftsmith explains. “For very factual tasks, like summarization, you want the output to be objective, not argumentative. For others, you may want it to take a stronger perspective. To get it right requires a lot of balancing behind the scenes.” 

Users can help shape Everlaw^AI Assistant’s drafts by completing additional fields, intentionally chosen by Everlaw’s engineering team to ensure a more targeted work product: areas of law, parties represented, key arguments, and points of emphasis. The work product can be based on selected pieces of evidence chosen just for that draft or all evidence within the Storybuilder Story.

Automatically Draft Timelines, Counterarguments, and More

Drafting assistance isn’t limited to statements of fact, however. Users can create custom tasks to cover nearly any evidence-based writing need. Set a custom task to answer a specific question – the impact of an event on a party, for example – and Everlaw^AI Assistant will create a draft answer in a matter of seconds. Just as with the statement of facts, a draft will include direct citations to supporting pieces of evidence. 

Everlaw^AI Assistant can be used to answer nearly any question in this manner. Ask the platform to create counterarguments to your case theory to help anticipate oppositional interpretations of the fact. Or put together a narrative timeline faster than any legal professional could do independently. Whatever your prompt, Everlaw will apply best practices to shape the input so that the AI output is primed for use within litigation or investigations. 

“Obviously, legal work requires an incredible amount of time reading and reviewing documents, and organizing them for use in depositions and trial,” Matthew Kaplan observes. “I hope legal professionals find that Everlaw^AI Assistant can not only help them more easily find truth and justice through their review process, but also allow them to do so much more effectively.”

Creating a Foundation Legal Professionals Can Trust

The potential for artificial intelligence in the legal industry is enormous. 

But its application needs to be responsible, considered, and well thought out, so that legal professionals can obtain the benefits of AI in a way that is reliable and verifiable, from a product they trust. 

That’s what drives the team of AI experts behind Everlaw^AI Assistant. And it’s an approach that is resonating with those in the profession. “The AI holds the promise of allowing the merits team to move directly from first-tier review to assembling information for use in depositions, motions mediation, or trial,” Gordon Calhoun, Partner and Chair of the Electronic Discovery, Information Management & Compliance Practice at Lewis Brisbois, explains. “When this happens within the Everlaw platform where we can verify it, I have confidence in its ultimate value. And, Everlaw Storybuilder, which provides a structure into which to install packets of information, will allow counsel to tease out relationships between data and issues that otherwise might not have been discernable.”

“The race to push generative AI into production is real. It is impressive to see Everlaw doing it quickly and thoughtfully, adding guardrails so legal teams can gain benefits while better navigating the pitfalls of gen AI,” says Ryan O’Leary, Research Director at IDC, the research and market intelligence firm. 

Three Ways to See It in Action

Everlaw^AI Assistant is launching in a small, closed beta. Those who want to experience it directly can join the beta waitlist here. 

Everlaw is also hosting an upcoming webinar showcase of Everlaw^AI Assistant, with a live demonstration and Q&A session. Register here to save your spot.

Finally, don’t forget to check out the product launch. Join Everlaw CEO and Founder AJ Shankar as he explains the technical origins of large language models that power today’s generative AI, see an overview of Everlaw^AI Assistant, and learn about the responsible way to implement generative AI in your organization with Shana Simmons, Everlaw’s Chief Legal Officer, and Patrick Hammon, partner at Pillsbury Winthrop Shaw Pittman LLP. Join the Everlaw^AI Assistant launch event here. 

Casey Sullivan is an attorney and writer based out of San Francisco, where he leads Everlaw’s content team. His writing on ediscovery and litigation has been read by thousands and cited by federal courts.