Hanzo's journey in AI actually began well before the recent unveiling of Spotlight AI last week and the announcement that Hanzo won the the LegalTechnology Breakthrough Award for best use of AI technology this week. When I joined Hanzo as a senior product manager in November of 2020, Hanzo already had two data scientists on staff who were charged mainly with internal R&D projects and working within our Chronicle product line to help with Hanzo’s QA automation.
As time progressed, our Data Science team consisting of Aidan Randle-Conde and Jimmie Weiss began to transition from R&D more toward feature creation and productization for Hanzo Illuminate. This transition led to two ML models focused on PII detection and user (conversational) behavior. Thus, our first Illuminate data science-based features were announced and shown at LegalWeek 2022.
Post LegalWeek of that year, we began to explore more product capabilities with our partner IBM. One immediate area of improvement allowed us to tap into IBM’s Watson Discovery and leverage their existing PII detection engine. We then continued adding to our ML capabilities by implementing the categorization of data content thereby helping our clients understand what users were actually discussing within large conversational data sets.
Then, as you may guess, something dramatic happened late in 2022. ChatGPT was launched into the wild. Not dissimilar from most companies out there, our product and data science teams immediately started discussing and contemplating where this technology might fit into our product line and our industry. Early versions of ChatGPT were not without issues, and by the time LegalWeek 2023 rolled around you couldn’t take five steps on the show floor without hearing someone talk about it. Ironically, during the show, a data breach at OpenAI was also announced thereby accelerating the discussions around data security and “guardrails” within the community.
Looking back, LegalWeek 2023 was a major turning point for Hanzo. The show has always been a roadmap for company directions, and certainly, 2023 was the start of something different. On the second day of the show, I went back to my hotel room and thought to myself, “We really need to start from the beginning and rethink everything.”
That night, I drafted several ideas and a few workflow diagrams on my tablet and the next day met with Julien Masanès, Hanzo’s CEO, to discuss the possibilities. Starting with the EDRM model, we methodically walked through the entire process and discussed each phase and where we thought AI might add the greatest value and provide the biggest impact. Comparing this against our own product, we concluded that one of our greatest strengths is as a hyper-effective, big data culling tool. Our discussion continually returned to what is perhaps the largest and most impactful part of the eDiscovery process and that involves decision making around content relevancy.
Post LegalWeek, with some rough ideas in mind, Hanzo kicked off an internal project called “Project Esquire.” Our team drafted an outline of the best potential ideas and the product, data science and engineering teams conducted brainstorming sessions and mapping out potential approaches. From “Project Esquire”, Spotlight AI was born.
Our teams coalesced around three core philosophies for Spotlight AI. First, after discussing data privacy issues that occurred during LegalWeek, we knew client data security was the single most important factor to consider when designing Spotlight AI.
Second, Spotlight AI needed to provide AI decision-making transparency. As with any new technology and specifically one where large language models and AI are involved, we knew that transparency would be of utmost importance to our clients. There were many lessons learned from other AI tools and technologies such as TAR.
Lastly, we wanted to design features that were practical and useful for our users. Generative AI and LLM’s are outstanding at many tasks and our team set out to create a simple, understandable process and feature sets that offer the ability to quickly inform users and help them concentrate on the most important and relevant information first.
With over 5 million messages collected for an average enterprise-sized chat-based matter, we designed Spotlight AI to understand the language of a user’s case description or their input so that Illuminate could automatically elevate the most relevant or potentially relevant content to the top of the data stack, allowing users to focus on the most important information first and foremost.
Once we settled on our core product tenants and mapped out the initial feature set, data science and engineering teams got to work. They crafted test datasets, comparison rules, and a rapid iteration UI that allows for substituting one LLM for another in a very short amount of time. Additionally, in parallel our team began researching available large language models, considering costs, speed and their flexibility, as well as the ability to rapidly tune models according to specific data sets.
It was during the early test-and-iterate phase that our partner IBM released watsonx-ai, a cloud-based platform that provides developers with access to a suite of AI services. Watsonx has since become a strategic part of our process by helping leverage their toolsets and expertise.
Last week’s announcement was the culmination of over 11 months worth of work. What began as a fairly basic LLM announcement last year has progressed through rough ideas and sketches culminating into what our team believes will be a game changer for the industry.
Spotlight AI may still be in its infancy, but the possibilities it holds are awe-inspiring. As someone who has witnessed technological breakthroughs in the legal/compliance/GRC tech space over the past 15 years, I believe we stand on the precipice of something truly extraordinary. This is a new era, and Spotlight AI is poised to drive positive change across our industries. The journey has just begun, and I can’t wait to see where the next few years take us.
Experience the first automated relevancy assessment engine to bring data insights in hours instead of weeks.