Workplace Search with Generative AI You Can Trust | AI in Action 2023

Hello everyone. My name is Shivam Trivedi and today I'm going to be showcasing our product clean. I'm a solutions architect here that has been here since our first paying customer, and I'm excited to show you how we can solve the problem of bringing enterprise search and generative AI into your workplace.

The core problem that Glean is trying to solve is that employees cannot find the information they need when they need it to do their jobs. In a typical organization, information is spread across multiple applications like Google Workspace, Microsoft 365, Atlassian, Salesforce, and a lot more. It's very hard to curate quality content to each employee since information is always changing and permissions are also always changing, the most pain is being felt by your experts, where they're constantly being interrupted from their workflows, uh, by employees who cannot find the information to do their jobs.

And as people work in different time zones, they work in different styles like hybrid, remote, and in person. This creates completely new challenges as well. This problem typically costs about 20% of your workday according to many studies that have been done by companies like McKinsey and Asana.

And what this results in is engineering teams having, uh, having to redo a lot of work because they are, uh, not able to find previous work that has been done on a specific topic and troubleshooting repeat bugs from scratch. Instead of like looking at previous resolutions, sales teams are onboarding very slowly because they cannot find the information they need on the accounts they on onboard, uh, on, on the accounts. They're onboarding upon, uh, support team members.

They have slow ticket resolutions time because they cannot find the documentation real time and may have to do multiple touch points with the customers as they try to look for information to solve the issue. And HR teams are obviously spending a lot of time, uh, curating content and that is a very expensive, uh, endeavor, especially in terms of time. Yeah, this is where Glean comes in.

GLE creates an enterprise search engine that is completely personalized to your organization and truly understands how your company speaks. And we use the latest advancements in adaptive AI and LLMs to achieve this. This search engine operates completely on top of your business data, so it truly understands like how your entire company operates and a permissions and, uh, aware and secure manner and can reference any content.

As it's generating content, this entire system is ready to go out of the box and a typical customer will spend less than a less than a week setting the system from scratch to the, to the production ready environment. The way GLE has built the product is in five layers. First, uh, we provide a completely managed infrastructure that can be hosted by GLE or by your organization, and we support multiple cloud providers, including GCP.

Next Clean provides out of the bus, a hundred plus out of the box connectors that are completely maintained by our team, and do a deep crawl of all of your common SaaS applications, including things like Atlassian, O 365, Google Workspace, uh, Salesforce, and more. On top of that, we have implemented a governance engine, which ensures that users only see what they have access to in the source applications, so your security and IT teams and continue with the privacy controls they haven't placed in the source applications without having to do any additional work in Clean. Next, using the content, the people context like team title, department, as well as the native activity of your users, we're able to build a deep enterprise knowledge graph so the system can truly understand what content is relevant to each user.

Last, we layer on generative AI at the top. So we can provide a gen AI interface, a j and AI interface on top of search that can provide you relevant permissions and force and completely personalized generated content. The result of this system is that engineer engineering teams can reuse components and find components that they've done worked on in the past much more quickly and reduce the amount of, uh, work they're doing over and over again, and they can locate documentation and previous resolutions a lot faster.

Sales teams can onboard on accounts a lot more quickly and support teams can really reduce the time to resolution for all the tickets that they're working on. Last, your HR team's, HR team is not spending, uh, a ton of time trying to curate content on an ongoing basis since our search is real time and can automatically detect what content is relevant to each user. When we talk about generative ai, predictable augmented generation is really the key to how we bring something like this to the enterprise.

Generative AI has four problems. One, it is not permissions and force. Two, it is not real time.

Three, it is not connected to your company data, and four, it tends to hallucinate or give you wrong information using something like retrieval. Augmented generation. You can fix all four of these problems since we use search for what it is good at, which is retrieving of information in a relevant manner.

And we use generative AI for what it is good at, which is reasoning, synthesizing, and summarizing of content. So here, what gling will do is that it'll search using a user's permissions and then passing the correct context into the context window of generative AI to generate the final answer. As you can see, so gene AI can understand the, understand the prompt, it can search using glean, it can select the content it wants to use to generate the content and then act and answer the final, final issue.

Now this fixes all, all four of the problems that I just talked about. Number one, since our search connects to our, uh, to your company data already, generative AI now have now has access to your company information. Our search is also a realtime search engine, which keeps track of like any permission changes as well as any content changes, which means the final generated content is also going to have the realtime context.

Number three, our search engine is permissions and post, so users only seeing content that they have access to, and as a result of that, the generated content is also permission sent force. So all of the content is only going to be something that a user has access to. And last, since, since search is already since search is already taking care of all of the relevance when it comes to content, the final generated content also tends to have high relevance to the user and, and only generates content based on the information that is passed into the context window.

So this means instead of giving you a wrong answer, it might give you a no answer, which is a much better scenario in an enterprise and organization. Last thing I'd like to highlight is that Lean is built on trust and it is site SOC 2, type two certified. It encrypts all of your data in the single tenant instance and it is also compliant with GDPR.

Lean's customers include some of the most heavily regulated companies out there, and we have customers across all verticals. With that, I would love to show you a demo of the GLE product in our live environment. So this is what Glean is, and I'm accessing Glean using our web interface.

As you can see, the GLE homepage, uh, has a lot of smarts built into it. So it includes things like your calendar, some announcements such your HR team might be pushing to you trending documents, uh, suggested documents and a lot more. But the true power of glean really resides in this search bar.

I'm going to go ahead and perform a generic search on a feature that I might like, like to learn a lot more about. Three things I'd like you to notice here. Number one, glean is able to understand company specific context like acronyms, synonyms, definitions, and a lot more.

Because this acronym means nothing outside of our organization and our search ranking can really pick up your company specific acronym, synonyms and definitions as well, since it's a, it's running in a single trend instance. Number two, all of the results that you see here are completely permission sent forth. So I am only seeing results that I have access to in the source applications.

If I do not have access to any app, any results, any documents in the source applications, I'm not going to see them in my search results. Number three, all of the search results here are completely personalized to me as a user since this is a signed in search experience. So the system truly understands my team title department, who I'm working with and what I'm working on.

Last, I'd like to showcase how we build a deep enterprise knowledge graph by linking documents across multiple applications. And as you can see, since this Confluence document is linked in multiple Jira documents, we can use this context and information mining to rank better. Sometimes it's not enough to find just the information you're looking for and you need to find experts within your organization who can answer, give you further help.

In this scenario, companies can use a expert detection feature where we can find experts for specific topics across your organization. And as you can see Phil here, because he's, he's written a lot of content around this specific topic is being flagged as an expert right at the top. In addition to that, if you're trying to look for a needle in a haystack document like something that you wrote a very long time ago, you can use a variety of filters at the top to find exactly what you're looking for.

With generative ai, things get a lot more interesting because I can ask pointed questions directly in the search bar. So let's say I'm an engineer who wants to understand how this feature works. As you can see, this triggers our assistant functionality and what this does is that it'll read through all of your searches results and then generate a pointed answer right at the top based on the question that you were asked in the search bar.

Since our search is already permissions and enforced and ranked and personalized to you, the result that the generated answer is also going to be exactly relevant to what you are looking for, and as you can see, it is also citing the documents that it used to generate the content. So this esta this allows you to establish additional additional trust so you know which part of the answer is coming from which document based on the citations we add. You can also ask follow up questions, uh, to glean, and this will trigger our chat interface where it is going to the do the same three things where it'll search using your permission, so read through documents and then generate a final answer.

Clean assistant can do anything that a typical generative AI system can do. It can do things like generate code on your behalf where I can ask it to do code work on my, uh, uh, code work for me on parts of product that I might not be familiar with. It can summarize calls for me.

It can also do like a lot, lot of research for me from Slack conversations and generate a document for me at the end. The possibilities are endless. In addition to that, glean also provides a platform where you can create context specific generative AI bots within your organization.

So imagine you want to create a IT support bot that can live within your organization to answer questions. So I can create a IT support bot or a help bot. I can give it access to the specific content that it is supposed to have access to, and then I can publish it as a Slack bot, an A PIA team spot, which is going to be coming soon or even embedded in a custom website.

Lean is providing a platform for organizations to bring generative AI safely into the enterprise. In this scenario, companies can, company admins can do all of the platform work once and then sanction the use of generative AI in a safe manner. Since all of the content is permission sent force and completely controlled by your organization, this demo should give you an idea of all the core features of Gleaned.

com and get click on the Get a Demo button and someone from my team will reach out to you.