Laws in Canada.

Is it two-party consent basically?

Always?

I think it might be one party.

Okay, well, I mean, I appreciate you asking, but yeah, go for it.

Yeah, amazing.

I just put it on there.

Yeah, it's one party in Canada.

Fascinating.

Yeah.

Zach, I just realized I just wanted to answer your questions that you put in the Slack as well before going.

Oh, yeah, sure.

I mean, we could do it.

We could do your first and then we can go to my questions.

That's fine.

Yeah.

Yeah.

Okay.

It was going to be, like, pretty quick anyway, to be honest.

That was, like, I could just tell you, and then we.

But, yeah, basically, we.

We're just testing with twilio.

Like, we.

It is working, but, like, just creating, like, all the docs around it and stuff, so that.

That's.

And.

And we do have outbound, like, at least, like, we've been testing.

That I'm able to put in an arbitrary phone number, make a call.

But we'll put some kind of more concrete stuff with that as we put it out there.

Do you mean the outbound part?

Yeah, outbound, outbound.

So I think if there's at least you're able to put in a phone number and make the call.

Yeah.

And like we're not, there's a few different use cases for that on our side.

Like one could be, there's a company that just launched, which I think it's kind of interesting, but like one of them is like chasing down physicians or doctors to sign a death certificate.

And like that's a pretty generic script you could probably give to a, to an agent, an agent and say like here, the doctor's name is Dr.

Smith.

This in calling in regards to so and so, like can you just confirm that they're going to get a death certificate?

It's like a friendly reminder.

That's a pretty easy one.

But other people have asked for an automated call to the family even that has a little script or something.

I don't think they're going to be full-fledged conversations.

It's almost like a voice-based reminder.

That's such an interesting one.

Would that call be the call that notifies the family or is it a fall, they know the relative has passed and it's a procedural call after that?

I think it's like a follow up and it's with the families that they've already talked to.

So it's not like calling a net new person.

That's probably not something we would do.

Yeah, yeah, totally.

That's interesting.

Yeah.

Okay, cool.

Yeah.

What stuff do you guys have?

Because I've already taken up.

Sorry, Zach.

I was just gonna say.

Oh, sorry, Jack.

I was gonna say on the SOC 2 stuff.

Oh, yeah, go ahead.

I think that's a Q4 thing, but we'll circle back with a more clear date as soon as we have one on that.

I know that was already something you were talking to Stuart about.

Okay, got it.

Yeah, I don't know what that means for us if I'm being honest, but we'll think through that when we get to that.

We just ended our type 2 observation period.

We're just waiting for the actual report.

And so anytime we launch with a new integration partner, we just need to verify that they have done all the right precautions.

But even if you guys were like in observation or something like that, that's probably good enough to be able to be like, okay, like we're working through it.

I'm assuming you guys are gonna do-.

Quickly as possible, yeah.

So I can, I'll check in with Erin on that, 'cause she's leading that.

I don't have the latest on where it is as of this call, but we can definitely circle back there.

I think, yeah.

So you in observation stage would be like the first checkpoint you guys would need.

I again, I don't know.

Like it would depend on what like we're using Delve, Delve.co, which is one of the AI security companies.

So presumably, like when I add you guys into our dashboard as an integrator, I just need to say, like, here's how it's secure essentially.

And it's only if we get audited.

But I don't think we're going to do an observation period until much later anyway.

So we might be okay.

I just like, if we talk about this, we have this big conference coming at the end of October, sort of like the dream force of the funeral world, which is just a fascinating experience.

If you're in Chicago in October and want to see a wild time, let us know.

But we

might, I don't know if we're going to launch anything by then, but we'll probably want to mention and have some sort of workable prototypes or demos that we can showcase there.

So as long as we're not actively launched, that's technically Q4.

So I think we'll be okay.

Okay.

Yeah, that's great.

Those are really good contexts and notes.

I'll relay that to Aaron.

We can kind of

have a more meaningful discussion there on that of exactly where we're at and those timelines, maybe once we get through the Twilio stuff.

I think that's going to be coming very soon.

Okay, great.

Okay.

Amazing.

Okay, Zach.

So, technical Advisory board.

So we'd love to just kind of really.

It's kind of like therapy, hopefully for, like, voice AI and, like, I know it's pretty early, but, um, all these questions are just, like, around building voice AI and your experiences and.

Challenges.

And I know, yeah, it's actually probably just the first question is like, have you started to prototype stuff at all?

Yeah.

Or like.

Yeah, yeah.

So we have, I think we first started with your hosted backend and just sort of prompt engineering.

What is the right balance of directions and step-by-step instructions?

Versus

we had tried other tools, right, where you're building full workflow based, where are you in the decision tree?

And we actually found that that's problematic because it's really easy to get stuck in a weird loop.

And then you're like, how do you have all these escape conditions?

It's really, really complex versus I think the models have gotten better with context windows where we can just have one prompt and it manages the whole thing.

We've started with that, but then once we knew we wanted to do something with Twilio and be able to extract information out of that, maybe do our own tool calling and that sort of thing, that's when we moved into, okay, we could switch this to our hosted backend and our server and then be able to handle that.

We have a version of that as well.

Then thanks to you guys launching Twilio, we have that.

There's a number I can call right now that will pretend to be one of our customers.

Receptionists and we've used that just to sort of gut check that we have the right building blocks.

I think the reason we haven't pushed forward on that is call recording because we know we just know we're going to need that.

So we've sort of waited for that and then the minute we do that, we'll continue with the prototyping process.

Okay, that's amazing.

Thank you for the context.

One very quick question just for my own interest.

Are you using any frameworks for like building the kind of agent side of it on your own backend?

I had asked you guys about this and I'm sort of torn because I don't know if we...

So the short answer is no.

We're literally using one of the examples you guys put together, but it's hosted on Cloud Run on GCP.

We have a ton of

Google credits that are actually expiring in like a month.

So I'm just like, I will blow through all of them if I need to.

So just using their various models.

But on their infrastructure.

So not an issue, right?

I think down the road, we might switch to some sort of framework if we want to do more structured tool calling and variable extraction and stuff like that.

But I don't know if our calls are so complicated that we actually need that just yet.

Okay.

That's a super helpful background.

Actually, one other thing, sorry, is in a lot of our calls, we've gone through a few different examples.

One reason or pro for using a framework It's like, I need to do eight different tool calls.

I need to do obituary lookup and create case tool and then edit case tool and all those sorts of things.

And I'm sort of like, I would rather load all the context it needs in memory.

So like tool call find obituary.

No, just load every obituary in the next two weeks into memory.

You know, that or any, any, you know, go a week back and go anything that hasn't happened yet.

Load that in the memory and that's fine.

And then for all the like, create a case, update a case, do this, do that.

I would almost just do that once the session's done, take the whole transcript, take the whole variable extraction, save that somewhere, and then do post-processing there versus doing a tool call.

That's my thought, because then all we have to optimize is latency and actually handling the call.

That's really, really interesting.

I don't know if that's right or wrong.

I'm just trying to think about simplifying that part of the code, right?

Yeah.

Yeah.

That should make the latency better for sure.

If it's like, you're like, all about to off so cool.

I don't want it to have to call anything.

Like, I just want to be able to, you know, it basically is the, the LLM and like what, what CPU and RAM that I put on the server, basically.

Yeah.

And I guess a human might do it that way as well, where they just make a note or something they need to do kind of.

Yeah, like the incumbent that I mentioned before, they'll take notes in their system, they'll print out, it's called first call sheets.

That's just essentially like a summary of everything they talked about.

Yeah, that makes sense.

And Zach, one question that we really like to ask, if there's anything about building with voice AI that you could wave a magic wand at, and make it better or make it exist,

what would you wave the magic wand at?

Good question.

I mean, this is a problem across the whole space, but probably to do with testing evaluation sets, things like that.

Because if I change the prompt ever so slightly, I want to make sure it doesn't break anything.

And so I know some of the other providers have the concept of running evaluation tests, which I know is less on the voice side.

It's basically just testing the text that the LLM is returning.

But even that, I think, would be huge if you guys could work something like that into the platform.

How would it change your life if we had that?

How would it change my life if we had that?

I think we could move a lot faster.

Because if I change the prompt, the only way I can test right now is by calling it or maybe just doing the LLM stuff on my side, like with that endpoint.

I could build all that myself if I want, but it would be great if it was somehow factored into this.

Just because I've seen it with so many other platforms.

Maybe you guys will never do it and that's fine too.

But if it's just the infrastructure, that's great.

But testing it is almost as important to me as the infra.

There's a company we're using called extend.ai.

It's for document processing, classification, extraction, all that sort of stuff.

If you log into their product and check it out, they do a really great job with evaluation sets.

That you can run against different versions of various processors and it'll show you

the diff of, hey, this increase, this decrease, that sort of thing.

We've been using that for part of our OCR product

for directors and that's been really, really helpful.

I'm not saying you need something exactly like that, but if I could say, hey, we're going from

this model to that model or this prompt to that prompt, how can I actually be sure that nothing's going to break?

Some weird edge case that we would love to test for isn't going to screw up.

Is there anything about having that that is more valuable now than it was like a year or two ago?

Do you mean like because the models are better or just in general?

I guess like just kind of I don't want to lead you like too much in it, but like it's like if you had to say like why this is more important Is there anything that like, this is more important today than it was like a year ago to be able to like.

Test what.

Test you're like,

yeah.

To test it, it isn't, I mean, it's, I don't think it's changed in the last year.

I just think it's very like, cause with the OCR stuff, for example, like if we get it wrong, they type it out anyway.

And it's the director that's dealing with it.

If we get this wrong here, a family chooses a different funeral home and has a really bad experience on the worst day of their life.

The stakes are just so much higher.

Wow.

Yeah.

Yeah.

No pressure for everybody.

Yeah.

When you put it like that, it's important.

Yeah.

And so like, you know, it's sort of like we use another company called Tusk and they have a bunch of blog posts about how they do evaluation tests against LLM based content or LLM based work.

And I just think for something like this, like I would want to write essentially unit tests or evaluation tests for like somebody calls and like they're threatening their own life.

Like let's have a bunch of tests for that.

Somebody calls and they're in full shock, you know, like we need a triage and immediately send that to a human, like all of those sorts of big, big edge cases.

And then also the like, They said a homophone.

Do they ask how to spell that?

You know, like basic things like that.

But I think that's just going to make this sort of thing way, way more.

Again, sorry.

We could build this ourselves, but if there's like some way that was baked into the infra, that would be game changing for us, I would say.

Yeah.

Yeah.

I

don't want to share prematurely, but we're hearing this a lot.

Is what I'll say is that this, what you describe as, like, what you would wave the magic wand at is, is actually coming up again and again.

So I think this is something that we're going to think a lot about.

Yeah.

I mean, the, the llm models are not deterministic.

They're, like, probabilistic.

And so, like, if everything is, you never know what you're gonna get, or it's, like, a little crazy, right?

So like then the surface area of.

Weird shit on voice systems in production seems to be like it can get quite big quite quickly, right?

Yeah.

And I don't know how, I mean, that's like beyond everyone in this call.

Like presumably that's the model.

Like we just, there needs to be like global prompting that prevent that.

I don't know the right solution there.

Yeah.

Zach, just a couple of questions about how are you learning about voice AI at the moment or in general, best practices, how to do things?

Admittedly, once we started working with you guys or exploring with you guys, I stopped exploring others.

But most of the learning was through looking at all of their features, comparing them, and then honestly just a ton of trial and error.

You know, and I'm, I'm only scratching the surface, right?

I've probably only made 100 test calls.

I probably need to make, like, 10, 000 to, like, really do it.

Right.

But other than that, it's just like, I still will occasionally look at Bland's website or retail's website, or, like, I'm on their mailing lists, I think, so I get, like, all of their product updates, you know, but.

That's pretty much it in terms of learning what else is out there.

Yeah.

Yeah.

Okay, that's really helpful.

Yeah, those were my core questions, actually.

Aiden, I don't know if you had anything.

No, I don't think so, yeah.

I think.

We'Re just thinking about what type of content is useful to people and what problems

you know, we can, we can help solve and guide people to solve.

And so that, you know, just knowing what you kind of pay attention to from a, I guess, like, what you're subscribed to is one thing.

But that, I guess, the, the follow-up question to that is, like, do you prefer newsletters, podcasts, videos?

Like, what should, what do you generally find is, like, the most helpful way if you're kind of seeking to learn more about how to solve the problem.

I'm just curious, media diet, I guess.

I'm more of a, I would rather read an email or a newsletter or an article.

I don't want to hear people talk about it in a podcast and I don't need to see a video.

Yeah.

My co-founder is laughing at me right now.

But like, you know, if we, if we, if we did a podcast, right, frankly be two white dudes talking about, hey, so I wanted to get this receptionist to take my restaurant order and I found out that if I did this prompt instead of that prompt, It used dim sum instead of dumpling.

Like, I'd be like, this is the dumbest podcast ever.

I need to get a podcast.

No, no, no, no, no.

What, what have you been doing a podcast for?

What was the initial idea?

Podcast.

Okay.

Can you guys hear me?

Hello.

Yeah.

Yeah.

How's it going?

Nice to meet you.

This is Jack and Adam.

Okay.

So they were asking, like, how do we get, news about voice agents and, how what he was asking me how I would best,

how I would want to consume that media diet was the phrase he used.

And I was like, absolutely not podcasts.

Oh, your ideas podcast, not them.

No, no, no.

He was saying as like, is it newsletters?

Is it podcast?

And I did to be clear, like, not specifically about voice AI, it was just generally in you, like, what do you, what's your preferred method for learning?

Right.

Reading or listening or watching?

And I think, oh, well, for, I.

Would say for technical topics, it is like written articles or tutorials.

I just think I've seen people do like live demos of tech in terms of how you do A versus B.

And I just feel like those are never like, I don't want to see you type into your terminal.

No, no, he's asking media diet and he gave some examples and I just said, no, no to podcasts.

I just don't think podcasts are helpful.

That's all.

Yeah.

Big thumbs down on video and audio.

Yeah.

I mean, obviously your mileage may vary.

I'm going to go back headphones.

For asking again for a.

They waved by, by the way.

I just think, yeah, I don't know.

Like, do other, are other engineers that you're talking to wanting videos or podcasts for this sort of thing?

I think some people.

I think the.

Value of written content for like a technical reference is just like unquestionable.

I think some people like watching tutorials for technical stuff, like developer tutorials.

That's like a type of video that people have said to us is helpful and I know people like.

And then I think the general kind of podcast stuff, like, I completely agree with you.

There's a lot of types of conversation, which is just like, it's not,

I would not want to be doing those, but I think Kind of an open question for us.

Who's interesting to chat with about what they're doing with voice?

I think maybe podcast conversations, to my preference, the more technical they get, the more you're into this should be an instructional piece of content or a doc.

But I think higher level, even just as an example, you guys talking about your business, it's a very interesting business.

You're solving some kind of novel problems with voice at a high level.

It's cool to hear about that kind of thing, I think, for maybe a conversation.

I think we're just thinking about the various levels of education, what opportunities are out there in voice.

And I think our team, we would rather put out less content and have it be really useful than try and do too much stuff.

So yeah, I'm just generally thinking about what is useful to the people that we're actually working with, because that's the best group of people I think we have in terms of

we already kind of know you a bit.

Obviously like what you're doing and are interested in it.

So yeah, that's where the question comes from.

Yeah, well, I think you kind of hit the nail on the head there, which is if it becomes more of like an instructional technical resource, then like a doc with a guide is better.

If it's like a, here's something cool that you can do with voice agents that you never thought of, we're going to interview the founders of so and so company.

Yeah.

Then a podcast makes a lot more sense because you kind of want to know their story and hear it in their own words.

But I think if you were to say something like, hey, we found out that if you change the turn timeout from X to Y, this is the end result in the customer experience.

I don't know.

It's like a headline.

That's like the headline of the blog post.

And then the blog post is like, here's an Roi.

Here's a net result.

And then here's how you change the setting right now.

Sign up for your layer code account, basically.

Right?

Like, I don't think it needs to be more.

Involved than that.

100%.

Yeah, I think my feeling on kind of like what we're seeing in voice is like there's two buckets of things that are really interesting.

One is like just unsolved technical problems everywhere.

That's like interesting in a slightly different way to like people building stuff you never would have thought could exist that involves talking to computers.

And sometimes there's some overlap, but yeah.

It's super interesting just to get that perspective from you because

we just honestly don't even have too much of a bias or preference on this show.

We're just thinking about where we should put our resources.

Yeah, I'm curious too around, do you have an email subscriber list or are you just going to be putting this on Twitter or LinkedIn or something?

Yeah, we have a marketing email list we've been building and just Even with building audiences on platforms, right?

It's like there's too many platforms.

Now we kind of have to pick, and I think LinkedIn, Twitter, useful.

Yeah, thinking about YouTube again, it's like lots of technical and structural tutorial type stuff plays well there and stuff.

But yeah, if we do ever send emails with this kind of content, I really want to make sure they're good.

We probably will.

But.

The anti-goal is like sending people stuff that is just another newsletter that they don't look at.

The bars.

Yeah, I would agree with that.

I mean, I mentioned the other ones that I'm still subscribed to for whatever reason.

I basically only open it and say like, are there any new features I should know about?

And then everything else is kind of thrown out the window for those companies specifically.

I was going to say, if you ever end up doing

a podcast around the companies, what you should do is have Stuart call on someone at Workday, and it's Workday, Stuart and me, and that way Workday will put it on LinkedIn too.

That's a good one.

Oh yeah.

If

you could stomach the pain of talking to people at Workday.

That's a good idea.

Yeah, you said it, not me.

Yeah.

I think there are two people from Workday that I would hire.

Everyone else, I wouldn't.

You'd get them.

There'd be two, like, there's two engineers that I worked with where I'd be like, I would work with them again, but everybody else is probably no.

I never worked with Stuart directly for the record, so I'm not talking about him.

No,

they've still have quite a big office here in Victoria from our acquisition.

That's right.

Yeah, 100 plus people.

Right.

Well, didn't.

Was.

Was.

I'm thinking of Juan.

Ono.

Was he in Victoria?

I'm not sure.

He was.

I've been off to my time.

He's in Vancouver.

He's in Vancouver.

Yeah.

I mainly am connected still with just like the OG Media Core group that came in with us.

But yeah.

I have one last question for you, Zach.

Sorry.

No, that's good.

We are thinking about

doing some fun stuff for the members of the Technical Advisory Board.

This is a swag question.

Is there any swag that is Worth having?

This is a good question.

I think 10 years ago, I'd be like, yeah, give me whatever you got.

But, like, I don't actually think I have a good answer right now.

I think most stuff is trash.

Like, it ends up in the trash or, like, recycle it.

So, like, why are you wasting your.

Your venture-backed money to just, like, buy people t-shirts or whatever?

I would only get so she's involved now.

I would only.

I would only.

I would suggest.

This is so unlike this cover.

Okay.

What I was going to say is if you're giving us something, I want it to be functional.

I don't want it to be like something I wear.

Like a coffee mug or like something that will be helpful to me as a developer even, or like, I don't know, something like that feels way more interesting to me.

Yeah.

That's great feedback.

That's great feedback.

For like for context, the only swag that we get now is

Funeral home swag.

I also have a nail file.

That's a funeral home swag.

Hilarious.

Oh, that's a good one.

Yeah.

Or we have.

Oh, that's good.

We have a small casket named Mort

and a embalming fluid machine.

So, yeah, that's.

Oh, my God.

Okay, that's what you're competing against.

Okay.

That's very helpful feedback.

But I would also say, if you're getting swag for everybody and I'm being a, I'm poo-pooing on everything, just give it to somebody else or donate it or something.

I don't need it.

Our feeling is the same as you.

I think we all have way too many boxes of startup t-shirts and shit we've never worn.

And it's like, if we're going to do something nice, we want to do something really nice.

So, yeah, we're just thinking of, well, we're asking a bunch of people that question and just thinking, what would be fun.

So thank you for.

Yeah, that's a good.

That's good input.

Yeah.

I remember I was so excited to get socks from slack, and then I was like, this is dumb.

I'm never wearing, like, they're not even good.

Like, why am I.

The worst swag I ever had was my workday mug when they did the onboarding with my employee number on it.

Oh, yeah.

Yeah.

What a terrible culture for that.

You are number

1017.

So I was like, this is awful.

Reduced to a number on day one.

And then you go in a meeting and somebody walks in and their mug is like 17 and everyone treats them like they're the word of God.

And I'm like, dude, you're not even building a language.

You're literally building a UI and calling that programming.

So can we not?

It was,

I'll end on this because I feel like you'll appreciate this if you didn't go through something similar with them.

The first thing they did when they acquired our company was come and paint all of the blue things, all of the red things in the office blue.

They sent a guy on like 10 days after the acquisition before anyone else came to Victoria because they don't like Oracle, right?

Like nothing can be read at work day.

And we kind of asked why they were doing it and I think they said something like, well, obviously you can't have read things in the office when the senior team comes out here.

It's just like, that is the weirdest piece of culture that I can remember.

It's a cult, yeah.

But like they all are.

I was gonna say, if you have to send us something.

We don't have to send you something.

Well, Effie is texting or messaging me right now, but her's idea is way better, which is don't give us swag, give us free credits.

That's what would make more sense.

Okay.

Give me the monetary value or some multiplier to whatever makes the most sense to just be like, okay, you get this much credits.

Which also for the record, I don't even think I've put a credit card into your system, probably because we're not using it fully, but I still feel like there's, yeah, I have $15.

I don't know how I have $15 in here when I've never added it, but do you give us credits?

Is that what it is?

We should top that up for you.

I'll top it up.

Yeah.

Is that equivalent to meeting?

She's a great negotiator.

This is a good bit.

Is this a bit?

I don't think this is a bit.

I think she's like, I will fight you, get me credit now.

Well, if you're down to $15, I feel like we should bump that up.

Yeah.

Cash strapped.

I recently started using blacksmith.sh.

It's a much faster GitHub action.

If you're familiar, but they have this cool thing, which is like, if you tweet about us or put it also on LinkedIn, we'll just give you like, 300 extra minutes of credits.

And you can do that every month.

So that's cool.

Yeah, that's so cool.

They have a really sick onboarding as well.

I've seen that.

It's like it was painless.

Yeah.

Check, check, check.

Done through that one.

Yeah.

Yeah.

Anyway, any other questions?

Because I've already taken up more time than probably.

No, this is.

Thank you for giving.

We've taken up your time.

This is amazingly helpful.

Yeah, well, no, this is great.

I'm really excited about the call recording stuff.

And then whenever, I mean, I'm happy to also be a guinea pig too on the outbound stuff, even if it's not even in the UI, if it's like pure API call or something, I'm totally down to test that.

And I don't necessarily think we would launch with that part anytime soon.

I just want to be able to show that we could potentially do it to figure out how we would do it.

Yeah,

go on.

Did you really can't hear?

We can't.

Is that funny?

I know that this isn't exactly part of it, but could we analyze a live call through them?

Like, if you guys hear that.

The phones of funeral homes, could we be like continuously analyzing what they're doing?

Because it'd be cool to pop up with them, like, this is this person's case, and we want to do that anyways.

So can we do that in real time?

We don't do that anytime soon.

I'm going to jump in here, sorry.

In real time, technically we could, technically we do that now because we're streaming the call content, right?

That would just be something on our endpoint.

Am I right in thinking that?

Yep.

You

can get the text back.

So then you could just, if you wanted to, yeah, you could basically just, you want to do an action based on what's happening in live time.

Yeah, I think so.

Yeah, you could do that totally.

Yeah, you both said totally.

Okay.

Now this is a bit.

Okay, anything else that I'm forgetting here?

Sorry, nothing on outside.

Okay, cool.

Well, yeah, I mean, let me know whenever those things are good to go.

And yeah, if you need anything else for me, just reach out or ping me in Slack.

Yeah.

Would it be okay to chat in about a month and we could share what we've picked up on the calls and like confirm if you think it sounds legit?

And then also we can give you updates on like the progress on this stuff.

Yeah.

You can if you want to do like one time meetings every time you're ready or if you even want to just do it recurring, that's totally fine.

Okay, great.

Yeah, I could see whatever's easier for you and we can always shift if.

I think we'll probably have a couple updates on these features before then.

Right.

Yeah.

Awesome.

I can send you a recurring link thing with Calendly.

Perfect.

Thanks so much.

All right, see you guys.

Get an umbrella.

I'll see you soon.

We'll try it.

Stay dry.

All right, bye.

See you.