Building Better Tech: A Look at Relativity's Partnership with Microsoft and Elasticsearch

Our mission at Relativity is to organize data, discover the truth, and act on it. It’s important that we do this well—and not just because one of our core values as a company is to exceed the expectations of our customers and colleagues. It’s important because the impact our users have on the world is incredible, and our job is to enable their success.

Thanks to the hard work of the people in our community, RelativityOne is plugged into not just career-defining M&A work or high-stakes litigation, but urgent investigations, compliance activities, data breach responses, and many other legally—and ethically—impactful use cases. The outcomes of these cases can change careers, industries, and lives.

And in today’s highly connected world, it would be impossible for us to build software that’s robust and extensible enough to manage all of this in a vacuum. Our vibrant user and partner community is extraordinarily helpful in bringing our mission to life. Some of those partners include Microsoft, Azure OpenAI, and Elasticsearch, and I had the privilege of working with them to put together some content for Microsoft’s Build conference this spring.

What is the Microsoft Build Conference—And Why Was Relativity There?

Hosted annually, Microsoft Build brings the developer community together with key Microsoft experts. The conference centers around the latest innovations in software development, coding, and related technologies, and offers breakout sessions, workshops, networking opportunities, and other ways to connect and learn. Imagine a version of Relativity Fest built specifically for our own developer community, and then make it Microsoft-sized—that’ll give you an idea of what I mean.

Build attendees include engineers, developers, and IT professionals who represent a spectrum of expertise and experience levels—and many of whom work for tech companies that use Microsoft platforms to build and enhance their own software. Relativity is one such company; we have a long history of partnership with Microsoft.

From hosting RelativityOne on Azure, to creating connectors that enable more efficient collection of Microsoft data types, to empowering users to translate documents on the fly with Cognitive Services, to collaborative artificial intelligence R&D, to cybersecurity mind sharing, and plenty more—well, our touchpoints with Microsoft are everywhere.

Thanks to our strong partnership, Microsoft thought of us when another one of our partners, Elasticsearch, expressed a desire to work with a customer use case as part of their previously scheduled Build sessions. It was a whirlwind of activity; In just over two weeks we created a use case, generated over 50,000 documents, built a working Elasticsearch environment, and crafted and delivered the presentation. This was at the same time as Relativity Fest London, so it was an exciting May!

Build itself was a blast. Microsoft puts on a heck of a conference. From the quality of keynote and session content to the production value and effort put into making every session a success (including professional hair and makeup—which, let me tell you, after flying directly from Romania to Seattle, was a gift from the universe), it’s clear that they are deeply committed to delivering an excellent customer experience. Overall: 10/10 would Build again.

A Universal User Conundrum: Explosive Data Growth and Diversity

During our session at Build, we talked about one of the biggest problems facing our customers right now: the data growth explosion, and the myriad file types and applications they may need to search through as they’re working through each case.

Compounding the problem is a natural language shift across applications; you probably don’t (I hope) email someone like you would Slack or text them. These changes in tone and style make keyword searching—once the cornerstone of the e-discovery process—a less effective mechanism for identifying important documents in a data set.

In Elasticsearch Relevance Engine (ESRE), we see the potential to deliver a search experience that goes beyond keywords and basic conceptual searches. This technology offers an opportunity to natively tailor the search experience to our users’ workflows, including customized indexing based on actual case data. We also see a search experience that’s augmented by numerous AI capabilities—think GPT-4 classification and prompt models and much more.

Our hope is that, by bringing together AI like this with the irreplaceable human intelligence of our users, we can help turn something that takes months of work today into something our users can achieve in days—or hours.

This is the type of innovation we’re working on with these partners, and it’s what we covered for attendees at Build. To put the technical into a more real-world perspective, let’s walk through a fictional example of how this near-term future of e-discovery searching might look.

A Future-of-Search [Fictional] Case Study: Investigating Labor Trafficking at BigThorium

Imagine you’re an investigator with the US Equal Employment Opportunity Commission. You’re presented with a case of two undocumented workers who were recently fired from their positions and subsequently deported. The case revolves around BigThorium (a fictitious energy conglomerate), which put out an international call for workers to build its new power plant in Texas.

The claimants in your case are welders who live overseas. They were contacted by a local recruiter and offered a spot on the BigThorium crew. However, they had to front the processing costs of their employment. Both sold their homes to scrape together $20,000 in fees and head to America.

A month later, the claimants found themselves relocated to flimsy barracks-style housing with their other countrymen. All of them were segregated from other workers by a fence and armed guards. And they had to pay exorbitant rent to live in these conditions.

Their recruiter had promised them green cards, but they ended up receiving temporary work visas. Obviously dissatisfied with the reality of this arrangement, they wanted to leave—but management withheld their passports, garnished their wages, and hinted at the possibility of physical violence should they continue to complain.

Ultimately, the claimants texted about the option of whistleblowing—without realizing BigThorium was listening in on their conversation. They were subsequently fired and deported.

As an EEOC investigator, your job is to get to the bottom of what happened and ensure justice is served.

At Relativity, our job is to make that work possible.

(Quick note here: this hypothetical case might sound despicable and outlandish, but our BigThorium scenario is modeled after a very real case currently being litigated by the ACLU. And these stories are depressingly common; see our documentary about labor trafficking for more information.)

Off the bat, you find yourself in possession of more than 50,000 documents that need to be whittled down to just those that are potentially relevant to this matter. How do you start?

Today, in real life, investigators often find themselves at the base of a mountain of data—with very few known facts, and little insight into where to find more of them. There are a thousand potential paths to take to begin that investigation.

With RelativityOne, we aim to provide some signposts that help identify the quickest, most direct path to the truth. We do that using technologies already built into the platform—like fit-for-purpose sentiment analysis—and we’re working with Elasticsearch and Microsoft to engineer even more of them using technologies like ESRE, GPT-4, and signal detection.

For example, in your BigThorium investigation, RelativityOne can help you interrogate the data with a question or natural language prompt—say, “conversations related to pay cuts.”

Right off the bat, when you type this out, the AI suggests adding “and workplace retaliation” to your query. And when you hit Search, the engine identifies some spot-on documents that offer a proverbial smoking gun.

Naturally, your workflows and results won’t always be so black and white—but it isn’t impossible, and you’d be surprised how often it really happens.

Putting Responsible AI First

This not-too-distant future is thrilling, and the possibilities of greater efficiency and shorter routes to the truth hold a lot of promise in the legal space.

But the use of AI is not without risks, and we’d be remiss to ignore that reality. Injecting AI into the legal process is a delicate job, and the questions of “how can we do this well?” and “how can we do this right?” are, deservedly, getting a lot of chatter lately.

In our work with our customers as well as with Microsoft and Elasticsearch, we’ve been adamant about keeping our ethical obligations at the forefront of our AI development strategy. We’re hyper aware of the responsibility that we, and our partners, carry to get this right. Enabling fairer outcomes in the justice system is no small task.

This is exactly why we’ve defined and published a set of principles that guide our decision-making and development around AI. This framework provides accountability within our organization, and ensures we never lose sight of what matters most: the people impacted by the insights our technology surfaces.

These principles are centered around fairness and accountability. Because it’s our obligation to provide our customers with security, privacy, and control over their data, and to build transparent, understandable, fit-for-purpose AI tools that can scale and treat all people fairly.

Put simply, we are committed to carefully considering the legal and ethical implications of everything we build—and to thinking about not just if we can build something, but if we should build it.

The Power of Partnership

We have our principles to guide us in how we craft new innovations for our customers, and thankfully, we have our development partners to help us bring those big ideas into the real world.

Technologies like ESRE and GPT-4 offer so much promise for legal applications of many kinds, and we’re excited to continue working with Microsoft, Azure OpenAI, and Elasticsearch to further explore those possibilities.

Many members of the Relativity team, myself included, have talked a lot recently about the need for diversity of thought in AI development to identify and reduce bias, and partnering with other technology companies is a great way to introduce that diversity of thought. The expertise that my co-presenters brought to bear on AI-augmented search, for example, was humbling to experience.

Most importantly, in my work with Elasticsearch and Microsoft, it’s been clear that they share our passion for responsibly building solutions and are just as concerned as Relativity when it comes to our legacy of technology on the world. Working with AI comes with a great deal of responsibility, and it requires thoughtfulness, consideration, and compassion to get it right. That’s doubly true when combining the forces of three major technology innovators and providers. If they didn’t share our passion and our drive for responsible AI, our work just wouldn’t be successful.

Closing Thoughts

The last several weeks have resulted in two major takeaways for me.

First, the future of searching in Relativity is a bright one, and the customer experience we can generate through these partnerships is something I’m very excited about. In a few years, it is my fervent hope that our customers will be searching in all new, much more efficient ways.

Second, our customers and user community deeply care about responsible AI. They are rooting for us to get it right, and they are supporting our careful and considerate approach to development. That means that we aren’t motivated by a frantic race to the top of AI mountain; instead, we want to thoughtfully find the best path there, even if it means it takes longer. The trust our customers are showing us by encouraging this approach is truly humbling. And I thank each and every one of you for being on this journey with us.

Elastic, Elasticsearch, Elasticsearch Relevance Engine, ESRE and associated marks are trademarks, logos or registered trademarks of Elasticsearch N.V. in the United States and other countries, used here with permission. 

The banner graphic for this post was created by Kael Rose.

---