Hi guys. Um, really nice to meet you all and um, thanks for First Mark for having me here today.

I'm Sneha, the co-founder and CEO at Spur and yeah, excited to kind of take you through the presentation today.

Just a quick agenda. I'm going to go over what Spur is, who we are, our customers, investors, and then really spend most of the session today on um giving you a demo and what we call spurring in time.

Spur is a new way to test.

Um we are basically building AI browser agents that automate software quality assurance with agentic AI.

I know that there's a good number of technical folks in the audience.

Um it's very long very hard to build super flaky and yeah it just doesn't really scale.

Now with spur the idea is a test on spur um you know you can kind of build it in 10 minutes. I say 10 minutes this could even be a minute.

The idea is you can write tests in natural language.

The inputs and the intent behind a test is ultimately the script that is given to an AI browser agent.

And we'll dive deeper into how this actually works under the hood.

So a little bit of background on the team. So Spur was co-founded by myself and my CTO Anushka.

Uh we both have a product and a research background.

So Anushka used to work at DeepMind and myself at Figma and we met at the Yale NLP lab.

We're backed by an amazing team and of you know a team of ex-founders, customers and competitors um as well as investors that you know we really love and we have an amazing team here and we're also hiring across um many different roles.

So short plug, please find me after the event if you're, you know, looking for applied AI, uh, you know, research, design, sales, etc. We're we're hiring across the board.

And we're working with some of the fastest growing brands in the world.

And you might wonder like why why brands, right?

So typically you saw the Selenium Cypress Playright test that I kind of showed earlier. brands and BTOC companies, they have user interfaces that are constantly changing, which makes the idea of having rigid tests just doesn't really scale and really bleeds the need to have a more agentic approach for testing.

So with some of our customers, the impact that we've created is really, you know, cutting down that release cycle time from one week to a couple of hours, from 6 months to even 5 days.

Um so it's it's pretty incredible what a gentic QA can do for um QA teams.

So the technology powering spur as I mentioned Anushka and myself we both met at the NLP lab.

So the underlying technology is really um browser agents and browser agents trained for specifically software testing and quality assurance tasks.

So under the hood it's powered by LLMs.

The idea is you can now give natural language instructions and instead of having to go through a step of making it into a test script in code, you give those instructions to a browser agent that goes in and executes those actions simulating an actual end user or customer.

So the idea is the agent can catch a lot more things than a typical script can and even what you and I can which we will I will be putting you guys to the test shortly.

So under the hood you know we support all different model providers and because of this you know from a testing perspective right we're going way beyond just functional testing.

Um we have a whole set of agents that we support from action agents, assertion agents, extraction agents, um UIUX testing agents and persona agents.

So essentially it's not just QA in the sense of does this flow work but also is this flow optimal?

Does it really feel good as an end user?

And then going a little deeper into the underlying architecture.

Um the idea is the enentic inputs as I mentioned can be natural language instructions but we take everything from you know a text input a Jira ticket you know a video or loom recording and actually we use Gemini's models for that um and or even like PDF or CSV files of test cases and then that gets into a system of robust robust multi- aent architecture where we have overseer agents kind of overseeing both action and assertion agents.

So this is just one example.

Um I kind of shared that we have many different agents.

So each agent actually has its own um custom under the hood architecture.

But that's really where the magic happens.

Okay. So um now I kind of want to walk us through a couple of examples on spur.

Okay. So I'm going to cover a bug that we found on Alo Yoga.

There is a bug on this screen. So it's not addressed.

So, is this something typically like a coded script can catch?

Probably not. It's impossible.

Like imagine on Black Friday, you're going on a sale.

The reviews don't even match up to the product that you're buying. Are you going to actually buy it?

Um, and this is millions of dollars lost for a brand like Alo Yoga that's doing 300 plus million dollars in revenue.

So, this is actually an example of a real bug that Spur caught for the customer.

This is an example of a test result on Spur.

The inputs for this test are actually on the left hand side of the screen.

So, as you can see, um the instructions are very dynamic.

So, I'm going to just zoom in a little bit.

If you can see some of the steps here, which is close the password page if present, then wait for a homepage to load.

The idea is these step these steps are exactly like you would be talking to another human.

So the way we tell our customers in terms of how they should think about writing tests is really taking a shift from how can I codify these steps to really like how can I prompt engineer the spur agent to QA my site in the way I wanted to.

So kind of going into the actual bug and error description.

This is typically the output of a test and particularly this step which is basically caught that the review summary that was extracted earlier refers to a dress but the current product is a 3-in high-waist shot shorts not 3 in I don't know what that three means but [laughter] um [clears throat] the specific review summary about dress versatility fall weddings and girls night out is not displayed on the page because it's not relevant to the shorts product being viewed and this error description is also generated by the agent.

So the customer can really trust the agent to catch everything that even you know even like a human tester might not see.

The idea is for any test that runs um we give users a video playback of the actual agent execution as well.

So I'm going to hop into the next example.

Okay, second bug challenge.

Um, this is a bug found on living spaces just yesterday. Um, right before their like code freeze before Black Friday.

On the left is the before and on the right is the after.

The bug is actually on the after image.

Yeah, exactly. Um, no storage.

Um, so again, this is kind of another example of an error that the agent finds where you've basically set up the filter for storage and in that validation step, the agent is basically flagging this as an error.

The last example, which is um on HelloFresh, the prompt for the agent is actually pretty broad.

Um, you know, we've not really given we're giving the agent a persona.

So, I'm not going to read the full persona for you, but basically, you are a efficient, inspired user. You want to get tasty, fresh, home-cooked meals on the table, etc.

This persona is something that their UX or research team would provide.

With this test, typically for a brand like HelloFresh, they get a lot of bugs just by the unhappy parts, right?

It's not always just about testing the happy parts.

So the prompt given here is actually explore the page and click through the other pages. Evaluate what you see, what you're missing, and suggestions for changes and warnings. Don't give up too fast.

So as you can see, this is a 25m minute long test.

So the agent is actually going back and forth while it's performing the execution.

And I'm going to just play the the bug that was found.

And this was on hellofresh.com US.

So you can just imagine how big this was.

So in this test the agent is actually going through the funnel and as you can tell um it's basically gone back a step and in that process of going back the entire page is actually blanked out and this is it you know this is a simple react state bug but you're not going to catch something like this if you test things purely linearly which again a coded script um is really not going to be able to give you that level of coverage.

So the idea is with these prompts, it's really up to the end user to kind of um be creative and you know come up with different use cases for the agent and kind of go beyond that functional um codified test approach.