Introduction

All right, hi, I'm Nana. I'm on the product team at this company called u.com. And since this is the practical talk, I'll be demoing a few tricks that are really great for research productivity using u.com. So I'll get started.

All right, so just a quick intro of u.com. So we are experts in providing accurate answers, first and foremost, to complex questions, and also we're helping you automate workflows to make you more productive. In short, we're trying to be the trusted place of Gen AI, no hallucinations, good citations, and get the answers that you actually need for more complex questions.

Just a quick background on the company. We're founded by Richard Socher and Brian McCann. They're both leading AI research scientists from Stanford and Salesforce.

And originally, I won't go through their whole timeline, but they were tackling personalized search. And then in 2022, you know, LLMs blew up. And so we connected essentially LLMs and the AI capabilities to internet search.

Exciting Features

And so today I want to show you two features that I'm personally very excited about. These are actually pretty hot off the press. So you guys will be getting a pretty new look into our product.

Advanced Source Controls

One is advanced source controls. So being able to kind of control what sources are going into your AI queries.

And then the second one is advanced research and reasoning, which I'll take you through in more detail. So since this is a product demo, let's see how it works in live. All right.

All right, so I'm just showing the home page of what u.com looks like. It might feel familiar to you in terms of there is a query bar.

Productivity Agents

And we also, I know agents was mentioned earlier, but we have made a bunch of agents kind of across the board of departments. You can kind of see marketing, sales, engineering that you guys should try out on u.com. We're really trying to kind of harness the productivity aspect of AI, and we'll show you how in a minute.

So here, these are already pre-made agents, but I'm going to go to More and Add New. And actually, before I get into this part, by a show of hands, have you guys used the Google Search Advanced Source Controls? Do you guys know what I'm talking about? Some of you? Yes? Yes? That sucks.

Or, okay, so how many of you have ever kind of queried something in one of these AI tools and thought, what are they sourcing from? Where is this kind of answer coming from? And what is this random blog that the answer is coming from? Anyone? Yeah, okay.

So we're trying to basically tackle that problem of having some control and reliability in your answers. So let's walk through it.

So I'm going to use the example where I've talked to a customer who is a founder of a biotech firm, but he does a lot of medical kind of research queries for his company. And so let's say I wanted to do medical research helper.

Creating Custom Agents

And I won't go through all of this today. Obviously, you can add prompt instructions. You can add description.

And one neat thing about u.com is we incorporate all of the LLMs. So it's kind of like take your pick. And we ingest all the new models. So whenever OpenAI or anyone drops a new model, we put it in right in. So you can just choose what base model you want to use.

And then this is the cool part about managed sources. So right now we're creating an agent and what we mean by like attaching a source is that we're kind of we're basically attaching a knowledge to and that becomes the context for for this agent.

And so there are ways to upload files, but I'm actually going to go to this web access, which is where the meat of the demo is today. So by default, u.com will just search the web for you and provide the most relevant sources. However, you might be a little bit pickier about what you want to put in your query. So we have this advanced search setting. And if you go here, there's a few things that you can work with.

So one is search depth, which is kind of getting at, in a sense, like how much of the context you're kind of pulling from the web. So for this use case, I might go to deep.

And then since I can never remember the URL for this, a lot of medical researchers use PubMed, right? So I can add a custom web source here and add it. And essentially, I'm kind of like telling it like prefer this source, right? So I really care about the PubMed medical articles.

And you can do either, you know, I only want PubMed, which in which case you kind of uncheck this, or you can let u.com also search over other additional sources. And then, you know, the other functionality you have is this particular person that I talked to, he was like, well, yeah, sometimes I get all these like WebMD results and I don't want to use WebMD as a medical researcher. So you can, you know, put in a bunch of URLs here to exclude. So like I do not want, you know, results coming out of this.

And so I don't know what this is. So you can play around with this. There's many ways to craft it, and it's a trial and error for what you want to do.

And today I'm not going to add instructions, but obviously you can tell it, like, always write me X, Y, Z. And I'll create it. if the internet works. Yes.

Um, so from here you can query, you know, whatever you want, which, you know, I was kind of testing out impact of something like that. And so what our product does is essentially it'll go through this workflow, and give you an answer. And the cool part of it is in the sources, you're really only restricting to the sources that you said you wanted to search over.

This might sound simple, but this is quite unique in the way that we've crafted our product. It just gives you a way to really tweak the types of outputs that you get. And I think I loaded it here.

And just so, you know, obviously when you're on u.com, you don't have to do it this way. But it's a cool comparison when I just ask the same question using our smart mode, which is kind of our basic everyday application. good for everyday tasks mode. And you get an answer, but you do start seeing maybe you don't want jptcp.com to be part of your sources. And so you definitely see a difference in the quality of the answer.

Advanced Research and Reasoning

The second part of it that I wanted to get into is kind of this advanced research and reasoning. And part of this, the impetus of this came from, we've had a research mode for years and that's one of our best liked products, I would say. And this goes to, how do we kind of make it less of a black box? How do we provide transparency when the LLMs are providing an answer for you?

And so especially when it comes to these either complex and or very open-ended questions. So you might see something like, say something, this is a little bit random, but I'm thinking about buying commercial real estate in North Carolina. Like, how should I think about it?

So if you use research mode, It's going to take a little bit, and I can come back to this. But we get this answer, and it's a nice, sure it looks like all the LLM responses that you might get. It's a little bit longer.

But the neat part is that if you look in the workflow, it tells you what the steps were that it went through. And these are kind of things that like I probably if I sat down for 20 minutes, I could have put out a list of all the things that I might have searched one by one. But it parallelizes the research for you and essentially kind of like just does a deep dive like all at once.

And so they're investing in the current state of commercial real estate, promising regions and cities, what types of commercial properties, legal regulatory, maybe like tax factors, whatnot. And so you can click into each of these as well. And then you see, it's almost like you ran this query like individually, but you kind of get it all at once and you're able to see the end state of this output is actually kind of pulling from all of these pieces. And then it just kind of lump sums, summarizes at the end.

And it also tells you how many sources we considered, 131 sources read by the machine. And yeah, like this is a, I think it's pretty neat in that, you know, there's kind of more rhyme and reason to how the answer is being generated.

Combining Features

And to combine the two, you can do something where I've basically made the similar, you know, source controlled agent, here, and I called it PubMed Agent. But you can also enable advanced research and reasoning then. So when I ran this, it does the same thing. And you're also controlling the sources that it's using. 1And you got to see every single step that it was going through.

And just to kind of further highlight like how other products that you might have seen have done it. ChatGPT, I put in the same question, right? Like I'm thinking about buying commercial real estate. And they give citations and they give kind of an answer here.

But one, you can't really tell it, maybe you don't like a new Don Realty as a source. And also it's a little bit unclear what was going behind the scenes of how it got to this point.

Comparison with Other Products

So yeah, and then let me go back to... So to summarize some of these new u.com features, I talked about the first two, advanced source controls, like letting you basically filter and tweak what sources you want the LLMs to search over in the internet. Research and reasoning, which is, you know, telling you what steps it took.

Summary of Features

And then the agent library is basically, you know, we've created a bunch of agents for you. But if you, let's say you're on a team plan, you can actually see all the teammates, agents and share it across in this agent library.

Upcoming Features

And then just a sneak peek of what's coming. We have projects, which is a nice way to organize all your chats and threads and the work and artifacts that you're making out of these kind of LLM answers.

Self-serve RAG. Not sure how familiar everyone in this room is with RAG, but essentially it's a way to circumvent the context window limitations that LLMs inherently have.

if you've heard of like 200,000, 128,000 for GPT-4-0 models, whatnot, this essentially lets you upload as many files as big as you want and you can kind of dump a ton of context and you'll still be able to have the relevant context window just the way that we're serving it by indexing and searching.

And then last but not least, we're continually adding integrations to make it easier for you to upload more contacts such as Google Drive.

Conclusion

So that's it for my presentation. Thank you again. My name is Nana. If you have any questions and feedback, my email is nana.u.com. It's very easy to remember.

And then for sales and partnerships, please go to sales.u.com. And then we actually have, I still don't know what this black or white box is, but there's a Cyber Week promo going on, so it's 50% off the annual. Plan which is probably the highest discount I've ever seen That's it