Slightly Caffeinated

TJ MIller (00:00)
Yo, welcome back to the slightly caffeinated podcast. I'm TJ Miller.

Chris Gmyr (00:04)
I'm Chris Gmyr

TJ MIller (00:05)
So, Chris, what's new in your world, man?

Chris Gmyr (00:08)
a lot of the a lot of the same. And just

TJ MIller (00:10)
Ha ha ha.

Chris Gmyr (00:11)
cranking on work projects, cranking on side projects, cranking on house projects, so many projects, so little time to do it. And just ripping tokens and burning

TJ MIller (00:21)
Yeah.

Chris Gmyr (00:22)
GitHub action costs and all

the things. So yeah, nothing nothing too crazy this week. Yeah, just more of the same and we'll talk about a little bit more of side projects later.

TJ MIller (00:33)
Yeah. Yeah, man. I'm you know, I'm in the the same boat, you know, picking away on the side projects, doing a lot at work right now, which is great. you know, this week was my my son's thirteenth birthday was yesterday. So it's crazy that I have a a teenager now. And

Chris Gmyr (00:52)
Mm-hmm.

TJ MIller (00:52)
you know, his this was his first week back at school too, and we've got his laser tag party tomorrow. So it's been

Chris Gmyr (00:59)
So you that'll be fun.

TJ MIller (01:00)
Yeah, he got his VR headset. I don't I don't think he's taken that thing off yet. We like I woke up at like eight thirty this morning and he was already up in the living room playing games. So that's

Chris Gmyr (01:10)
Nice.

TJ MIller (01:11)
that's been pretty cool. I mean and and it was really awesome to be able to give that to him 'cause he's he's been talking about it since his last birthday.

You know, one of the reasons why we didn't get one for him last year was because doing some research, there was a lot of game restrictions based on age. And like that kind of unlocked at thirteen. So, you know, now that he's thirteen, he can like

easily play these games and not have to like juggle around like parent accounts and approving things and all of that. So it was it was really cool for him to be able to like open that up after like pining for it for a year and be able to to like deliver on that. So that was really cool.

Chris Gmyr (01:48)
Yeah. Nice. Well, that's sweet. And I'm definitely jealous about Blazortag tomorrow. That'll be fun.

TJ MIller (01:54)
yeah, it's gonna be a riot, man. I you know, we we got we bought a pack for up to ten kids, but I think the way it's shaking out is I think there's only gonna be like maybe three of his friends there. Cause it the the thing we didn't plan on was that it was over Labor Day weekend. I don't think either of us really noticed that. And so a

Chris Gmyr (02:13)
yeah.

TJ MIller (02:13)
handful of his friends are like doing things for Labor Day weekend, so like they weren't available to come.

but whatever, you know, whoever makes it is gonna make it and we're gonna have a blast, so it'll be fun.

Chris Gmyr (02:24)
Yeah.

Yeah, just grab some random kids in the parking lot or whatever. Hey, you wanna play laser tag with us? It'll be fun.

TJ MIller (02:28)
Yeah, for sure. Yeah, yeah, I'm assuming

there'll probably be like other people there playing too and so it'll be plenty fine.

Chris Gmyr (02:37)
Yeah.

Yeah, totally. There's usually like a big group and you're just a little slice of it.

TJ MIller (02:42)
Yep, yep, yep. So yeah. touching on coffee stuff. I'm I'm still moving through this Kirkland and it's it's not it's not so bad, but I'm I'm definitely pining for some really good coffee. We had a new coffee shop like open up nearby. Like I don't I'm trying to remember the the name of it. It's like

It it's a chain. and it's like drive through and outdoor seating, so it's kinda like a walk up booth thing. And I guess it's also like kind of a trendy thing right now. I don't I don't know I I wish I remembered what it was called, but we've been eyeing that up to go check that out and you know, try some try some coffee from there. So yeah, it kinda it

Chris Gmyr (03:25)
Nice. That's pretty cool.

TJ MIller (03:26)
popped up out of nowhere too, like

I don't even remember seeing it under construction and then all of a sudden I drive past it and it's like open. Like, where did this come from?

Chris Gmyr (03:35)
Yeah.

Nice.

TJ MIller (03:37)
Yeah.

Chris Gmyr (03:38)
well, one thing that I mentioned before and actually opened it up is the new honey processed one from

TJ MIller (03:43)
Yeah.

Chris Gmyr (03:44)
Purity. It's called Star Day. and I know I sent you picture of the packaging, but it's like has all like glittery like stars on it and almost like

TJ MIller (03:53)
yeah.

Chris Gmyr (03:53)
a like purple pink like galaxy on the side. It was just super cool. I'm like, TJ's gonna love this.

TJ MIller (03:59)
yeah, that's that

that was totally my vibe. A hundred percent.

Chris Gmyr (04:02)
yeah, and it's a a medium roast, like I said, honey processed, and yeah, it's been pretty decent. it says the notes are like honeysuckle, lemongrass, and apricot. don't get all of those, but it's definitely like a nice cup of coffee. So yeah, I'll put it in the show notes if anyone wants to check it out.

TJ MIller (04:19)
Yeah, it sounds so good. my gosh. And yeah, the packaging's right up my alley. So it'd be an instant buy for me anyway.

Chris Gmyr (04:27)
Yeah. Just

get it for that, like empty out the bag, put it up on your wall or something.

TJ MIller (04:30)
Yep, a hundred percent. Like I still I think yeah, over here next to my desk I still have my like artisan blend bag from when I picked that that stuff up from the the terminal

Chris Gmyr (04:40)
Yeah.

TJ MIller (04:41)
guys.

Chris Gmyr (04:42)
Yeah. Are they still doing that? I haven't heard or checked on them.

TJ MIller (04:44)
They're definitely still doing

the coffee. I don't know, Artisan Blend if that's still around at all.

Chris Gmyr (04:50)
Yeah. Cool.

TJ MIller (04:52)
Yeah, they're

I think I got the whole like terminal based website and everything. Yeah, I I I'm I think it's still around.

Chris Gmyr (04:58)
Cool. Sweet.

TJ MIller (04:59)
Yeah.

Chris Gmyr (05:00)
So yeah. Moving on from coffee. We were on another podcast earlier this week and not this podcast.

TJ MIller (05:06)
Yeah.

Yeah, that was super fun. I always love getting together with y'all. And I think I I think we should make a regular thing out of it, like a a maybe monthly, like a monthly round table podcast with all of us. That'd be that'd be super cool. So

For those who didn't catch it, it was on the North Meet South podcast. so it was Jake Bennett, Michael Durinda, David Hempel, and both of us. And we were talking about like Jake's been running into this issue where he's got you know, outside non, I'm gonna finger quotes like non-technical. It's mostly Jordan, friend of the show.

He is a technical person, but he is the COO, so he's in a non-technical role. And he's been doing a lot of like vibe coding contributions, like meaningful contributions of like reports or discrete pieces of functionality. And Jake's like, well, how do we vet this stuff? Right. Like it's

Chris Gmyr (06:05)
Yeah.

TJ MIller (06:06)
all coming in, it's like a lot to it's a big burden to review, and kind of talking about like how.

How to handle this, how to like handle the reviewing, maybe how do we sandbox things. and what I think what was super interesting for me is how they're handling it at Zillow, which is what David Hempel talked about. And they're they basically expose to the non-technical people like lovable accounts so that they can prototype these things and then hand these prototypes off to the technical team to then go and implement after it's kind of like

proved its worth and maybe been iterated on a little bit. And you know, I I think that's such a such a great idea to like get that concept out there, be able to get like feedback on it, have concrete conversations, and you know, really kind of tease out what the feature actually needs to be.

Chris Gmyr (06:57)
Mm-hmm.

TJ MIller (06:57)
So I thought that was a really, really smart way of approaching it.

Chris Gmyr (07:01)
Yeah, I really like that solution too. As long as you're able to use some of these systems, like we're we're not able to use really like anything out

TJ MIller (07:09)
Yeah.

Chris Gmyr (07:10)
of like bedrock for various reasons. So but if like other companies are, you know, okay with it or have you know different compliance rules and depending on the data that you can or cannot get into, then yeah, a lot of those tools would be super helpful for stuff

TJ MIller (07:28)
Mm-hmm.

Chris Gmyr (07:28)
like this.

So yeah, it was a really good discussion. So I'll put the link in the show notes and you all can check it out there. But yeah, I would love if we did like a a monthly or every other month and I don't know, maybe swap platforms or something like that. Like one time they come over here, next time we go over there and kind of go back and forth with that for different topics. I think that'd be pretty cool.

TJ MIller (07:46)
Yeah. I mean, yeah,

I think if we really want to get consistent about it, I could also see us like spinning up a a a podcast just for for doing the roundtables, you know, that was I used to do that with Eric Van Johnson and Sarah Gollum, with the PHP round table. And I think we were doing that monthly and that was a ton of fun. so yeah, I think I would I would love to make a regular thing out of it for sure.

Chris Gmyr (08:08)
Yeah. Yeah. Totally. So if you wanna hear more about that, and want us to actually put something on the calendar or whatever, let us know. any of the the five of us just ping us and we'll try and put something together coming up soon.

TJ MIller (08:21)
Yeah,

yeah, for sure. So this

Chris Gmyr (08:24)
It's

TJ MIller (08:25)
this next topic, man, I'm super excited about. you've shared with me some of the progress on it. you shared the website with me, which I thought turned out so good. so yeah, I'll let you run with this.

Chris Gmyr (08:35)
Yeah, so I've been cranking on another project outside of Hive, which I've also continued to work on. this one I was talking to my wife and she is in a local book club that meets about once a month and like the management is just kind of a mess. everything is going through, you know, like reply all email, sign up genius with ads and just lots of back and forth and who's doing what and

different roles like you know who's who's leading the discussion, who's you know picking the book type of thing. I'm like, man, this sounds ripe for an app or website

TJ MIller (09:13)
Yeah.

Chris Gmyr (09:14)
to do the management of it. So doing some like ideation with her and then kind of dumping that all into Claude, making a a an initial PRD and some like designs and stuff like that. I've been working on what's called novelhood.

So like you're reading novels in the neighborhood. we got super lucky with the name. My wife, after like iterating with Claude a bunch, like we didn't really like any of the names. And she's a like English major back in college and was a English teacher. So she's all about the words and the combos and stuff like that. She's like, I don't really like any of these. And she was thinking for maybe about an hour, she's like, novelhood. It's like perfect.

And then I did a domain search and we got the dot com and I think

TJ MIller (09:59)
Ooh.

Chris Gmyr (10:00)
the dot app too. So it's like super, I feel like just super rare that you find something like that. Yeah.

TJ MIller (10:05)
Yeah, do it, trust me, yes.

Chris Gmyr (10:08)
so yeah, I've just been cranking on that. the website is live. I'll put it in the the show notes. right now it's just a public landing page, but there is like a wait list. So if you are, you know, running a local book club or want to start one up or even doing like a virtual one.

It handles like virtual meetings as well. you

TJ MIller (10:26)
Yeah.

Chris Gmyr (10:27)
can sign up for the wait list. And then behind the scenes, I'd say I mean it's pretty functional now. I'm just like adding a a couple more bells and whistles and features. and then I gotta put like a super admin kind of above everything so that I can

TJ MIller (10:40)
Sure.

Chris Gmyr (10:40)
manage stuff. But yeah, it's been super cool. I've been like doing a bu a bunch of iterations on the application design, just making it a little bit more

you know, usable, modern, you know, nice kind of touches. but it's been really interesting because if you think about the people who do any sort of book clubs, it could be from twenty year olds to eighty year olds, you know? And how do you

TJ MIller (11:02)
Yeah, a hundred percent.

Chris Gmyr (11:04)
how do you make an app function, you know, that widely, like across the board? So a lot of it is gonna be

at least for right now, like email based like notifications and things like that. There's gonna be like a weekly digest of, you know, things going out. So hopefully it'll be easier to read, like easier to maintain, for people who are a little bit older. And then once you dive into the web app, you can do basically anything that you want in there. You like you can sign up for meetings, you can RSVP, yes, no, maybe. If you are going, you can have like additional guests up to like a cap for the room or the meeting location. There's like

privacy baked in if you're meeting at like someone's residence versus like a coffee shop versus you know like a a bookstore or something like that. you can do polls for the next book that we're gonna read, the next date that we want to meet at, meeting cadence. So you can set up like I just want you to build out the next year of meetings on like the first Thursday of the month. You can do all that stuff. and there's a ton of other things already built in there.

And then there's a bunch of design elements that like if you are older and need some help like reading, you can go into the settings and you can like adjust the zoom and the contrast and the design a little bit, which that's all kind of super cool for accessibility that I haven't really

TJ MIller (12:15)
Yeah.

Chris Gmyr (12:15)
done before. just again get the wide breath of people who potentially will be using that.

TJ MIller (12:21)
Yeah, man, that sounds so cool. I I'm not a reader. Like I've never been a reader. It is like pulling teeth to get me to like read anything. So I think something that like could be interesting here too is like maybe also accounting for like audiobooks or something like that too. Like I

Chris Gmyr (12:39)
Mm-hmm.

TJ MIller (12:39)
mean, maybe I don't know if there's any like any direct tie-in with the application that makes sense, but like I I would love to be able to like participate in

a book club where like I you know, people could read if they want or like there's the audiobook, like you can kind of work through it that way. that'd be really cool. And what what kind of also reminded me of all this, like a kind of a callback. Like you were pretty involved in the Curology book club too, weren't you?

Chris Gmyr (13:04)
I was we did a engineering like weekly book club type of thing where like we pick a book. We had like a whole database of like options, like notion database of options that we could pick from. We did like a little like voting like poll initially and then, you know, someone ran it for, you know, a chapter or two like per week. and that was super fun. So yeah, I could definitely see it being used for, you know, at work or, you know,

TJ MIller (13:27)
Yeah.

Chris Gmyr (13:27)
any any sort of book club like across the board. It doesn't really matter.

TJ MIller (13:31)
Yeah, it's so interesting, like seeing some of that experience. I'm sure scenarios that you you ran into there like also kind of informed what you're building now. So it's it's fun to just kind of like find those loops and callbacks to things. So yeah, that's cool, man.

Chris Gmyr (13:46)
Yeah. Yeah.

TJ MIller (13:48)
I think this is a great idea. And I love I love that this is like also something you get to kind of like build and do with your wife as well.

I've been I've been telling my wife, I'm like, dude, like now more than ever, like hit me up with your app ideas. Like you've always got like interesting ideas that pop up and I'm like, now with where AI is at, like we could make it happen. Like we could totally make it like heck, you could almost make it happen on your own. Like I could get you set up with like

a cloud instance and a layer val base and clawed code and like you could just go to town on it and like make your own web app, you know, and I could kinda help bump that along. So I think it's just like a really cool opportunity that you guys get to like work on that together.

Chris Gmyr (14:30)
Yeah, yeah. It's definitely more like heavily on my side because she's never done anything with an

TJ MIller (14:35)
sure.

Chris Gmyr (14:36)
app or whatever. But it's been really cool to work with Claude and Claude Code,

TJ MIller (14:39)
Mm-hmm.

Chris Gmyr (14:40)
especially with like redesigning different things and having the the shareable artifacts. because Cloud Code can make them now and it shows up in the web UI. Then I can

TJ MIller (14:51)
Yeah.

Chris Gmyr (14:51)
share it publicly and then send her the link and she can be like, Yes, no, I like this better.

you know, this feel type of thing. She sends that back to me. I turn off the the public share, so it doesn't get like indexed and all that stuff. and then I go back to Claude and be like, hey, here's the feedback that we want to make and like just iterate on that. And like

TJ MIller (15:11)
Yeah.

Chris Gmyr (15:11)
it's just been like kind of crankin' the background, which has been super cool. and yeah, and there's like book there's a book search in there. So you type in a title, it'll bring up the the cover, you can assign it to like

TJ MIller (15:22)
Yo.

Chris Gmyr (15:23)
a a meeting

You can have like those as like upcoming like polls or like reading up next type of thing. and it does yeah, it doesn't really matter if it's like the book or the audio book, like you can choose whatever.

TJ MIller (15:35)
Yeah, that's sick, man.

Chris Gmyr (15:36)
yeah, and then you can put in like different like segments. So if you're not finishing a book like each meeting, so like we did in the engineering book club, you can say, you know, this meeting is for this book and chapters one and two.

Then the next meeting is the same book for three and four and so on and so forth. or you do like you could do like multiple books. If you have like two small books and you meet every quarter or something like that, you can attach

TJ MIller (16:00)
Mm-hmm.

Chris Gmyr (16:01)
multiple books to it or multiple books with multiple segments. So it's pretty flexible. and this ties into things like open library and eventually it's gonna be tied into like bookstore.org, which is like an online local bookstore.

retailer. So what you could do is put some like options in there to say like, hey, this is my local bookstore that we want to promote through our club. And ideally want people to buy books through there through book club or bookshop.org. so that's all gonna be like coming soon, but it's just a different way to kind of support local bookstores and

Libraries you know, stuff like that too. So yeah, just been cranking on the ideas and

TJ MIller (16:44)
Yeah.

Chris Gmyr (16:45)
all sorts of things. 'cause there's a couple like options out there for book club like book club management, but I don't know. They're just not what I want.

TJ MIller (16:55)
Yeah, for sure.

now for the design and all of that, have you been using the the new Claude design stuff or have you been just like feeding it inspiration and letting it just kinda like iterate on that or?

Chris Gmyr (17:07)
It's been really hard to find good inspiration for again the the age range that I'm looking for because I love your stuff, but it's like way too like

TJ MIller (17:18)
It is.

Chris Gmyr (17:19)
techie out there, like crazy colors, which I love

TJ MIller (17:21)
Hundred percent.

Chris Gmyr (17:22)
which I love for you in your products. I I love them, but like thinking about like the broad spectrum of like 20 to 80,

TJ MIller (17:29)
Yes.

Chris Gmyr (17:30)
it's like, this can't be like super like in your face type of thing.

So I did a couple of like iterations. I went on dribble, got a few things that like got us started, but I w really wasn't in love with anything. So I installed a bunch of design skills. So not only like the built-in design skill, the front-end designer. also I did like impeccable, there's like a mill design skill, there's things like an Apple design skill, there's a taste skill, there's a whole bunch of like design type things out there.

And basically having to prompt Claude a bunch of like, hey, figure out what skill to do like multiple mockups and I want like three different completely different variations of like what you know about our feature set and where we want to go and design towards that. so it took a little while to I think we're on like the V three or V four of like the app redesign 'cause it's like, this isn't really, you know, working out initially. So

being able to spend some time with the different skills and it's like, I'm gonna use this skill for this and then like an animation one for that and kinda piecing it all together and doing multiple reviews and, you know, having me give feedback kinda along the way. So it was basically like cranking on the like updated redesign yesterday, which took basically like all day and then to like overnight last night. So it turned through a lot of work and tokens, but

the results of the the application once you're logged into it is so much better. So it's a lot more modern, but it's still like very friendly to any sort of age range. It doesn't go too crazy like either way. And it almost looks like different, you know, bookshelves are what you would expect, you know, scrolling through like a Amazon or a Barnes and Noble or like it's very book forward and kind of like timeline of dates that are like upcoming and things like that. So yeah.

Just super happy with the design so far.

TJ MIller (19:20)
Yeah, dude, that's really cool. I love, I love that. Like, you know, I'm with with Halcyon, I'm like, I'm totally cool with that. Like with the the the bright the bright design and the the like actual apps UI. I'm probably gonna introduce themes at some point, kind of like bear notes, to kind of be able like allow people to adjust and customize that. But I'm you know, with Sonari.

the the transcription app, I'm almost kicking myself a little bit for going so loud in the design because I wonder I I wonder if it's if if it's a little bit too much, you know, for for people who are maybe looking to use it like professionally, because it is all local. So like I could see like a doctor's office or therapist using this to you know transcribe notes or something and

It would be great for that, but I don't know if they're gonna like look at the design and be like turned turned off by that and be like, this is like too loud and it doesn't look professional and you know it's it's a very slick design, but it is very vaporwave, loud colors. and so I'm almost wondering if that is like that could like knock against it because the the big selling point for Senari is, you know, it's it's private, it's all local.

So it's it it suits those use cases of like doctors and therapists and stuff like that, like very well because nothing's leaving your system. But I yeah, I I wonder I wonder how the loud design is gonna, you know, maybe play into the play into that. So I know. That's that was

It's that's something that I've been starting to think about. And I think that that was a really good call on your end of like, look, I'm gonna be trying to serve this like really large range of different types of people and like how do I make a design accessible and approachable to everybody? And I like I love that the the color palette fits that, the website definitely shows that off. so I I think you're really like nailing that direction and it's been making me kind of think a little bit about how how I'm approaching that too.

Chris Gmyr (21:14)
Thanks. Yeah. and I think with both the apps, like you said, introducing themes and even starting, you know, with the default, like just commer or, you know, Mac OS, you know, type of theme. 'Cause I think the functionality of it is fantastic. Like the the UI itself is great. So just being able to theme that a little bit and be like

you know, I have maybe two or three different ones. So you still can keep like the super loud one and then you have more of like a tone down one for, you know, doctors, lawyers, whatever. You know, if

TJ MIller (21:45)
Yeah.

Chris Gmyr (21:46)
they don't wanna, you know, have fun with the application.

TJ MIller (21:50)
Yeah, for sure, man. no, it's so cool. I can't I can't wait to see this get out there. Like I think you're doing such a good job with it.

Chris Gmyr (21:57)
Thanks. Yeah. So like I said, wait list is open and I'm hoping to get some like beta testers in sooner rather than later, maybe within the next couple of weeks or so. So if you're interested, you know, just sign up and I'll edge you and reach out when when it's time.

TJ MIller (22:12)
Yeah, that's sick, man.

Chris Gmyr (22:14)
Sweet. So yeah. yeah, do you wanna talk more about scenari a little bit?

TJ MIller (22:18)
Yeah, so scenario's coming along really well. I'm super happy with where it's at. right now I'm deep in trying to build out the update pipeline. So like when I publish a new release, you know, the app will be like, hey, updates available and like handle all the like over-the-air updates and stuff. Now with Halcyon, this was super easy because I built Halcyon on Tori.

So it has all the affordances for handling updates and the update pipeline and all of that stuff. So that was really easy to build for Halcyon, but now Sonari is just it's straight Swift, right? Like I kind of am regretting not reaching for Tori, but Swift felt like the right play here. And it's mostly a met it's a menu bar app too. So

I I went all Swift with it. And so update pipeline is actually a bit more complicated. There's this solution called Sparkle, which is kind of funny, because cause that's what Prism was originally named. But yeah, so Sparkle is like the update pipeline for Swift apps. there's several out there, but this seems to be the most predominant one.

so I'm working on getting that integrated. And then once I get that done, I can integrate into the Swift app, but I've got like certificates and signing and stuff I gotta figure out. And yeah, so it's it's kind of a messy process, but my goal with Senari is hopefully to launch like two weeks, which gives me enough time to kind of refine it. I'm hoping early next week to hit up the

The mailing list and pull in a few more beta testers. the other thing that I'm working on is then like license keys and all of that stuff and how I want to handle that. right now I'm planning on using Polar for that. because I use that for Iris. It was super easy to set up. they handle all the license key stuff, the license key validation, distribution, payment processing. They they do they do it all.

But a friend of mine is also working on another solution that's very similar to that that I'm gonna maybe evaluate as well. but so that's that's the next piece is like licensing and everything. but functionally, yeah. Yeah.

Chris Gmyr (24:26)
Yeah, that sounds like a pain. Like thing and all that. Yeah.

TJ MIller (24:31)
So I saved the most pain in the butt stuff for the end, like licensing and update pipeline.

But for the most part, the app is like fairly feature complete. I'm trying to keep the feature set like very tight and focused. but I had a really interesting conversation this morning. somebody reached out to me on Twitter and sent me a DM. They have a sibling who has got some disabilities. Like I think they had a brain injury, paraplegic. and so they have some like speech impediment issues.

And they reach out to like, hey, do you think like Sonari would, you know, work for this use case? because they've tried Siri, they've tried could like I think they had listed like another two other solutions and they're like, none of them really works. There's he's got some like slurred speech, and I'm like, well, you know, Sonari's got a custom dictionary, so you could like kind of see what is being output and then add to the custom dictionary to see if that kind of fixes it, but I just don't think that goes far enough.

So I went I sat down with Claude this morning and kind of brainstormed it. And it seems like without a ton of effort, we could use one of the whisper kit models and do some Laura fine-tuning on it. And so I could maybe give him a let me yeet a cat here real quick. we could maybe give him a script, do some recording of, you know.

especially things that he's having difficulty with, and then do some training on the Whisper Kit model. And then, you know, be able to have you know, maybe somewhere inside of Scenari where you could load and assign like a custom local model. And it's just this is like a really interesting a really interesting use case for it. And I think, you know, I've I've spent geez, I think

probably 15 years now working with a service dog organization out of Pennsylvania. And my wife's done a bunch of work with them. We've gotten several of our dogs from them. you know, and so they I with that I work with a lot of like disability stuff through them too. And so this is just like a really it's a special thing for me to be able to like

provide some assistance or or like enable people who are having difficulties with things. So this is definitely something I want to chase down and see if we can find a a good solution for them because man, that would just feel like so good to be able to like have have that available and maybe that opens up the opportunity for many other people that we could like, I could put together some sort of automated training where people could come in and like load up, you know, have have a script.

have the recording upload it to somewhere in Sonari's back end, automate the the fine tuning of the model, and then they can just like sideload that into Sonari and have like a custom custom model for their use case. so it's just got all sorts of gears turning.

Chris Gmyr (27:23)
Yeah, I I'm loving.

Yeah, I'm loving this idea. And I think Scenari itself would be great for just like you and me using it, doctors, lawyers, whatever, fantastic. But almost like a scenario like Pro or whatever that has all this stuff baked into it and like you can train directly through the application because, you know, your disability might not be exactly the same as like my disability and the the way

TJ MIller (27:50)
Exactly.

Chris Gmyr (27:51)
to like

have everything be customized for the user and updated over time. So maybe some people, you know, just unfortunately get worse over time and the model needs to be updated with that. Hopefully people get, you know, some PT training and some other things going on and then it might be better over time. So I think the option to update the model constantly and consistently over time will also help the results.

from the model and scenari and everything else that the person needs to do on the computer with the dictation and and whatnot. So I think having that being like a constant kind of churning thing in the background, like, hey, you haven't updated the model in, you know, three months. Do you need any updates or how are things working? I think that would be like a super cool avenue that you could take it to open up the a whole new world for

people with these like disabilities and looking for some of these tools to like help them out on a daily basis.

TJ MIller (28:50)
Yeah, and one of the things with Scenarii too is we hold on to a certain amount of history that you can like go in through a history window inside of Scenari and like retranscribe something with a different model. And so I wonder if I do like maybe an extended application for these specific use cases, if there's that affordance of being able to like leverage history then to like

participate like maybe pull history in and maybe corrections inside of those history to then like auto retrain the models so you could say like yeah use this history and auto like auto train. that could be really cool. Yeah, all sorts of gear starter.

Chris Gmyr (29:26)
Yeah. Now, would you be able to do

like all that stuff locally or would you have to offload it somewhere, which is a little too tricky for privacy?

TJ MIller (29:33)
So look, I've got a I've

got a I've got a maxed out MacBook Pro, right? Like it's I got maximum RAM. It's you know, I I I tricked this thing out because I wanted it to last forever. And I wanted to be able to do stuff with local models specifically. So for me and my machine, I could probably do that locally. But who knows what kind of hardware other people are gonna have. So I'm thinking that this training system maybe

is like a back end piece of scenari on like a a Laravel app or something. And then that spins up a Q job that goes to you know, maybe spins up a fly dot Io container temporarily that has like

you know, a GPU associated with it or something that's like specifically designed for the fine tuning so that like you could upload this stuff to the website. I could kick off a VM or something that does the training and comes back. You could download your model and then load it into Sonari. 'cause I just like I don't know what kind of hard maybe maybe they only have MacBook Air or something, which is there's no way you could do post training on that at all.

So I I foresee it being something external to do that, but it seems just so feasible. Like I'm I got the message and like instantly had one or two ways I think in my head where I could like architecturally pull this off. And it is just like it I have such a soft spot for for this kind of thing that like I'm I think I'm gonna I think I'm gonna pursue this pretty hard and and you know hopefully be able to open up

that opportunity for many other people to have, you know, accessibility this way.

Chris Gmyr (31:12)
Yeah, yeah, a hundred percent. That'd be great. Sounds like a

TJ MIller (31:15)
Yeah.

Chris Gmyr (31:16)
fantastic venture.

TJ MIller (31:17)
Yeah. And so I like let me get scenarios out there. Like I said, I'm I'm hopefully on like a two-week timeline to go to to go loud and like actually offer this up for sale. I'm gonna do like early entry pricing. So for some promotional period, I'm gonna do a reduced rate and then you know we'll we'll take it up to the full price. so yeah, and I think after we get to that point.

th this is where I might turn my sights to with it and and kind of see what we can do because gosh, that would feel so good if I could enable, you know, enhance the people's lives. That would just mean the world to me.

Chris Gmyr (31:51)
Yeah, totally. Sweet. Yeah, and great great tips in there with like Tari versus just raw Swift, 'cause I've been thinking about a different Mac app that I wanna do. So I think maybe going Tori is the better way to go. But I have I have not broke ground on that yet, so

TJ MIller (32:09)
Yeah, yeah.

Tory, you know, and that's I I think you know, solo term Aaron Francis, he's talked a lot about why he chose Tory because I I a thousand percent agree with him that like frameworks are the future and Tory is a framework. It's got

standards and practices and taste and like affordances all over the place. And you know, it it made building out Halcyon really easy. And now I'm kind of finding those pain points when I did just do Swift. So it's like, all right, like that, you know, that makes a bunch of sense. I'm too far down the road to switch now. But yeah, like I I would definitely recommend going the Tory route.

Chris Gmyr (32:51)
Yeah, nice. I'll look into that whenever I'm ready to break around

TJ MIller (32:56)
Yeah.

Chris Gmyr (32:57)
on yet another project. But yeah, that'd be cool.

TJ MIller (33:01)
Yeah, so I wanna close the loop on something from our last episode. We made some predictions about Claude's limits. They were coming up on earlier this week on Monday, the extra 50% usage, right? That was expiring on Monday. And what happened? They

Chris Gmyr (33:19)
What happened

is they they did something a little odd. They said that they were going away, but not completely, and did a little bit of like AI math, which I don't really quite understand how they got to it. But essentially

TJ MIller (33:33)
They they fumbled this

announcement so poorly. Like they they did

Chris Gmyr (33:37)
Yeah.

TJ MIller (33:38)
such a disservice to themselves.

Chris Gmyr (33:40)
Yeah. so they're they said that they were getting rid of the fifty percent increase, but dropping it what like twenty-five percent or something,

TJ MIller (33:49)
Yeah.

Chris Gmyr (33:50)
but that made it mean that it was only like a an actual like seventeen percent decrease on what you have right now. So it was just very confusing. but they also extended the fifty percent to the thirteenth now. Yeah, September thirteenth. So we still have the fifty percent.

It's unclear to me yet again if they're gonna extend this even more and just kinda kicking the can down the road or solidify the the limits as they are, you know, right now or actually change them as what they promoted. But yeah, very, very confusing.

TJ MIller (34:25)
No, and and they

they could not have made it more murky. Like it was so funny. They had the announcement and then I think like an hour later it was like community noted of like, you know, of of like actually it's a fifteen percent decrease. Like someone had to come in and like actually get a community note on it that like made sense of everything that they put out because it was just so confusing. Like I don't I don't understand it did they have Claude do the math? Is like that what happened? Like

Chris Gmyr (34:55)
Yeah, I don't I don't who knows what they did.

TJ MIller (34:57)
man. So yeah, so th that's kind of what I predicted was that like they were gonna end up extending it again because how are you gonna continue to compete when open AI is out here just like handing out resets left and right and like way, way more limits. And they're like I think open AI even like clapped back on them with like, you know what, we're getting rid of the the five hour limit completely for anything over the basic tier.

So like there isn't even a five hour limit anymore for open AI's models. Like, I think every yeah, I think everything above the basic tier, they got rid of the five hour limit. So like how you gotta s you gotta stay competitive with this stuff. And like I I've already seen so many posts.

on on Twitter of people leaving, like canceling their Claude accounts, going over to OpenAI, because at this point, I think they released Astra yesterday, which is like open A's open AI's new frontier model that is supposed to compete with Fable 5 and 5.1. So like now we've got that model out here too. Like so I think I think Claude's got to figure out what what they're gonna do here to to stay competitive in this.

Chris Gmyr (36:09)
Yeah, yeah. but I love the competition between all of them now and it seems like everything is pushing in the right direction to give more back to the people and the users who have, you know, signed up for these expensive plans. And I there's always gonna be some shifts of, you know, pissing people off and I'm gonna cancel this subscription to go over to this other one. But

TJ MIller (36:31)
Yeah. It's what what are you gonna do? Like I I I I just I just wish they wouldn't I wish they would have just came out and said what it was instead of like being so weird with how they worded it. It like it made it feel like extra bad, like like you guys are trying to like pull one over on us. I don't know, man. Just it doesn't feel good.

Chris Gmyr (36:53)
Yeah, totally. but we did get five point one models for fable.

TJ MIller (36:57)
Yeah, yeah, we did. And so I I Halcyon Sync is still running. Like I started it on the twenty-second of August, and it is still going on Halcyon Sync. because every like I I asked it yesterday, I'm like, what is going on? Are we just churning or are we making progress? And it's like, well, this issue actually exposed four more things, and then this one we fixed that, and then there that actually found

Like when it did review, it found these three things. And yesterday I think it came back and even said like only half the spec was actually implemented before it went into rounds of review. And I'm like, what is going on? but once 5.1 came out, I immediately like I I use Hive. And so I went to the Hive lead and I'm like, look, as soon as we get to a stopping point, pause everything. I'm gonna relaunch this with a five point one lead.

and that has actually felt a lot better than than five.

Output's been so much less verbose, so much less confusing. So having 5-1 be the lead and then spinning up, you know, Opus, Sonnet agents is needed, and then having Codec still come in and doing review. that's been going great. And yeah, it it still says we're like three or four more days out from being done with with the sync engine. So

Chris Gmyr (38:18)
Always

always three or four more days out.

TJ MIller (38:20)
I'm just I'm crossing my fingers that when it's done it actually works because like she's it's just been cranking, but it it's it says like look, we've closed 17 lanes. Like it it's it is closing out work, but it seems like when it reviews stuff, it's it's either finding stuff, you know, gaps, things that weren't implemented, or it's like, well, we've it this feature

This piece of sync did get done, but you know, while we're reviewing that, we also found these other issues. And so, like, work is just spinning up more work at this point. I'm I'm gonna give it through Sunday, and like if we don't if I don't see real meaningful progress, like I gotta really dig in and figure out what the heck is going on and what I'm gonna do at this point.

Chris Gmyr (39:08)
Yeah, yeah. And that's what I've

been noticing too and trying to dial it in more, but it's it's hard to put that kind of more human flavor and like touch into it. Because when like you and I are making a feature, you know, old school by hand, it's like, well, let's put this off until later because it's really not needed to actually get the thing out the door and then we can iterate on it later. Like, yes, there might be like a bug or some sort of

sideways to get into this to break something, but we're really only working on the happy path right now. And let's get that out the door and then patch things as we get more feedback. And it's very hard to like ingrain that decision making process in like all these agents and systems and skills and and things like that, which I've been really kind of struggling with too. Cause it works sometimes with some things, but other times it goes completely off the rails and it's like, we need 20 more tickets to

Do this. It's like, no, I'm trying to drain the backlog, not add more to the backlog to get through this. So it's been a lot of trial and error and still not super successful with trying to dial that in so that I can fully be like autonomous.

TJ MIller (40:18)
Yeah, and you know, it's I don't know, man. It's it's so frustrating, but like it told me like, Look, your standard that you set forth at the beginning of this is you wanted bomb proof sync, and that's what we're going after. I'm like

All right. I I mean I guess I did set a high standard with that of like I want no data loss. I want, you know, just this bomb-proof sink engine. And so I guess I maybe set the standard too high, but you know, whatever.

Chris Gmyr (40:48)
Yeah.

TJ MIller (40:49)
yeah. So on that note, you want to wrap up?

Chris Gmyr (40:52)
Wrap

TJ MIller (40:52)
All right. Thank you all so much for listening to the Slightly Caffeinated podcast. Show notes, including links and social channels, are down below and are also available at Slightly Caffeinated.fm. If you have any questions for us or content suggestions, go to the Ask a Question page on our site and we will feature it on an upcoming episode. Thank you all so much for listening. We'll catch you next week.