Introduction

Hello, everyone. Please excuse my nerves. I normally do this in a phone booth on a call, so this is probably the biggest crowd I've done this to, but we'll run with it.

So maybe to kick off and sort of set the context a little bit on Sarno and what we do and what we're working towards. So we were founded in 2016. We are a Stockholm-based company, and really the goal of Sarno is to solve the meta problem of learning and access to information.

So really the problem specifically that this is trying to solve is access to data and information and then being able to take actions upon that as well afterwards once you've found that information. So anyone who's worked in a company of a certain size or anyone who works in sort of the knowledge working space knows this is an issue.

The Problem of Information Access

There's a McKinsey study that says that Professional service workers, so your consultants, lawyers of the world, waste 9.3 hours a week trying to find information. And so that's something that we're really trying to tackle.

Can you all see my screen OK?

Is this too zoomed out? Maybe if I come in a bit. Excuse the UI breaking a little bit.

The Solution: Integrating Tech Stack

1So at a fundamental level, basically the way that we do this is we integrate with your entire tech stack. So think of your email, calendar, CRM, ERP, any old record system that you've got from 20 odd years ago that the company refuses to depreciate, we will connect with it. As a user or as an admin, you can do that.

For example, if I come to Dropbox here, I can just connect to the organization or connect for myself. It's all no code. Anyone can do it.

And then once we've got all our integrations set up, we can then start searching, querying, and chatting with all of our company's data. So you can see these are the sources that I use most commonly. I've got my Notion, my email, Salesforce, Google Drive, and so on.

Demonstration of Sana's Capabilities

To kick off and set the context a bit more, I'll do one quick sort of simple prompt just to show you how it works. So now I'm asking, actually bear with me, make sure I'm in the right instance. Yep. Oh dear. The hotspots died on me. I promise this isn't Sana. Right, we're back.

So if I want to ask it quickly, I'm just going to ask, what is Sana's InfoSec policy, just to show you how it works. So chat GPT style interfaces I'm sure most of you are familiar with. But now what it's doing is it's searching across my entire company's record base and information that we have.

And so you'll see that it's given me the answer, and it's also given me some sources of where this information came from. Now, I can click into these if I want to. It'll go straight to the line or the page that that came from.

Or I can also use Sana as just a pass-through to the world's most powerful LMs. So we are LLM agnostic, we have the Meta, OpenAI, Mistral, Arma, Claude, and so on, and you can pick which model you use. And we have basically, using the large language models and our own custom web search, you're able to search the web and do research as well.

So probably, as you've also seen, similar to other LLMs, I can ask, for example, what was NVIDIA's revenue in 2023, and what are their biggest risks for 2025? Nothing crazy new, nothing groundbreaking here, I guess, also passed through to the LLMs. But what becomes really powerful is obviously the amalgamation of your company's knowledge and the large language models and everything else that sits in the platform.

Advanced Features

So, the next piece that I really want to show you is our meeting note taker. Now, it sits in calls, it summarises, it transcribes, as you've probably seen with other meeting note takers, but what becomes really powerful is then that meeting becomes a knowledge file that I can search and query and chat against.

So you'll see we've got the video here, we've got the transcription of what it was. It's given us a quick sort of key topics and action items, and this sits in all my calls every day. It's probably the most useful thing in my job.

But now I can ask questions to it, and I can say, for example, what were the biggest pains that Acme faced? And so I'm doing this for one call, but you can imagine obviously doing this across multiple calls simultaneously. This is where it becomes really quite powerful.

Now that we've done this, again, I can see the inline sources, inline citations, and I can see where it's picked up Max saying this in particular. This is one side of it, obviously. As I mentioned, the focus really is the information retrieval and then the actions upon that information.

A lot of the time in roles, we have a lot of repetitive actions, things that we do over and over again, and likely they'll be prompts that you send to large language models over and over again. So we have what we call tasks. 1These are the repetitive things that you do day in, day out in the role.

These can be, for example, run a competitor analysis. This is one that I've made. I need to get my usage up. It's quite low.

This one I'll show you for the call specifically. So for those of you who are fortunate enough to not have worked in sales before, Banked is effectively a sales framework for looking at the viability of a deal. And so as a salesperson, you probably have sort of six, seven, eight calls like this a day.

And so now what I can do is I can take that original call that I just showed you where I was able to ask questions to it, And I can just find this again. I'll pull this source in. And now what you're seeing here is the pre-written prompt. This is basically saying to the LLM, or saying to Sana and then the LLM, go into the meeting, find all this bank criteria, and update Salesforce for me.

So now when I run this task, you'll see that it's gone from pulling this information from the meeting that we just had earlier, it's found the record in Salesforce, and now it's going to automatically populate and update it for me. So this is just one of the things that we connect with, you know, CRMs, ERPs, ticketing systems, whatever it is. But again, it's about the speed in which you can find information and then take that action.

So I'm now gonna update Salesforce. And now just actually, just to show you, I'll jump in quickly and show you where it's gone. Anyone who's used Salesforce before knows this is a horrible experience. But you can see that it's come down here, and it's now added this.

You can see 75,000 Freya and I initiatives, and it's got it from here as well. So if you think about what's happened there, we've gone from a call in which someone probably had to sit and take notes. Maybe they missed something. Maybe stuff slipped through the cracks.

They're not really engaged, not really asking the right questions because they're scribbling notes. They then had to synthesize those notes, understand them, pull out the right bits, update the record system in a separate tool. all of which, excluding the actual time spent in the call, probably takes, I don't know, 20, 30 minutes. And in what has happened in the sort of two minutes I've been speaking and showing you this particular feature now. So again, it's really about trying to both give you access to knowledge and information across the company, but also help you do real work with that, you know, and focus on the mission-critical things and the actual things that push the company or the role or whatever you're doing forwards, rather than wasting your time trying to find information.

Handling Multiple Calls and Documents

Now, this obviously works for one call, but what if you have 20, 30, 40 calls? What if you have 20, 30, 40 documents that you want to analyze simultaneously?

So what we've built for this is Sheets. You can see Sheets as sort of like a Google Sheet, but with ChatGPT built in, if that makes sense, or an Excel with ChatGPT built in.

So now all I'm going to do is I'm going to give it a prompt and say, Again, all going through Sana to the LLMs. And I'm going to upload five or six 10K annual reports. And I'm going to say, from these 10Ks, I want to look at revenue, profit slash loss, risks, and who are the board members.

So now Sana is contextualizing this. It's understanding what I'm looking for. And it's basically put all of these things as different columns in this table.

Now I'm going to browse the content, and I'm just going to pull in a little folder I have of annual reports. If I just show you quickly, for each of these, it's created an individual prompt that's being sent to the LLMs. For every single item on this table, it's running an individual query. For every single cell, you can see a cross-reference between the source input and then the question that it's asking.

We are contextualizing and understanding the document, and it's called vectorization, basically turning it into a mathematical representation. giving that extra context to the query that we then send to the LLM. So SANA doesn't own, doesn't train any LLMs. We just work with them and give the LLMs the, how can I say it, the supplemented information to answer your query more effectively, more accurately.

So again, if I jump in here to this first one, you can see it's pulled out this number for NVIDIA. They made 60.9 billion, well done NVIDIA, in 2024. Again, I can see inline where this came from. Actually, if I click into this just to show you, it's going to jump to the line and the page that it came from.

This is quite a big document to load. There we are. It's jumped to the page it came from, Revenue for Fiscal Year.

I know I keep sort of reiterating on this point, but the idea is that, again, it's sort of easy access to information, and then what are the sort of critical analysis and what are the important things that you can do to actually get relevant things from that data and that information, because that is the most sort of fundamental part about it. And in this instance, we've just done it with 10Ks, but you can put meetings here, you can put any files here, you can put different data sources, you can say, look at these different SharePoint folders, whatever it may be.

Conclusion

There's a lot more to show in the platform. I appreciate I've only got 15 minutes, but it is free. We have a free version and an enterprise version.

If you're on enterprise, come say hi. I'll be at the back by the pepperoni. But feel free to try the free version as well.

It's just sana.ai. And I'd recommend that you connect your calendar. It'll start joining your meetings.

And yeah, I hope you got some news from this and got an idea of what can be done in more of an enterprise focus. And thank you for your time. Oh, and thank you to Josh as well and the MindStone folks.