Samuel's video 🚑

so we'd like to present Scarlett which is your personal analytics expert and live voice presentation agent.

So the idea behind Scarlett is that you can give the agent your connection string, ideally read only to your database and you can chat about your data using natural language using your voice.

So right now through a common API we can support Postgres, DuckDB, BigQuery, Snowflake, Redshift, all of these.

So no matter where your data lives you'll be able to use this tool.

We think that Scarlett is especially cool because you can get going in minutes.

You just give it your read only connection string and the rest feels very intuitive.

The agent is also able to generate interactive charts built on top of Apache Echarts.

So within the browser, within the chat you can actually edit the chart, change the range of data and better understand it.

And finally as mentioned this is completely done in natural language.

We do have text available but you can chat using your microphone.

All right, let's jump straight into using Scarlett.

You can sign in either with GitHub, Google, or a custom credential.

We'll just use Google for now.

And when you sign in, you're met with a simple chat interface.

All you need to do is enter your database connection string and then ask Scarlett whatever you want to.

Hey, Scarlett, can you give me a quick analysis of my database?

Perfect.

And now you're ready to submit your query.

Once you submit your query, Scarlett spends some time thinking about how to answer it.

It starts by introspecting your database schema.

it figures out what queries it needs to ask, submits those and processes the results, and then generates some charts and analysis for you.

For this demo, we're using Northwind, an open source data set that contains some example commerce data.

You can see the Scar has done a pretty good job of giving us a, quick analysis.

Started with a pie chart showing us, how our products are distributed by category.

It then told us where all our customers are located and it even gave us some information, about our orders.

Over time, we also get some analysis for each of these charts that gives some further insight.

All these charts are also interactive, so if, for example, we want to focus on data from 1996, we can just go ahead and shrink this down to what we want to focus on.

Once you've taken a look at what, Scar's told you, you can continue the conversation, by asking some more questions.

So once you're happy with the analysis that Scarlett has done and you're happy with the context in the chat, you can actually view the presentation that Scarlett generated.

So here, let's click that and let's see what we have over here.

We have a nice slide editing or presentation editing interface so that you can see what the LLM has generated.

So basically, we pass in the entire context of the chat and it automatically pulls the data, it automatically pulls the charts out, into the presentation and generates bullet points and a script to follow.

If you're not happy with what it did, you can change anything you'd like.

Once you've edited the presentation and are happy with it, you can go beyond just creating the slideshow and actually use, text to speech to present it, verbally.

Now, one other thing we've included is the ability to change languages.

So while I only understand English, my high school French is not that great.

If I were presenting to someone who only speaks French, I could change the language here.

That would allow you, to use an LLM to automatically translate it, and use Boson's system to play back the presentation in the language you've selected.

Once you're happy with the presentation you've created and you've selected your language, you can go ahead and click Play, which will start automatically translating your presentation to the selected language, as well as creating an audio transcription of it.

So now let's listen to a presentation class Montrome.