Controversial AI Gadgets vs Build Your Own (You can too!)

0:00 This is Memoad Gem, the lightest AI wristband that remembers everything. This is my old Apple Watch that was just laying around with a noteaker that I built. Both can record, turn it into

0:11 notes, and remember everything. All the comments I read about AI gadgets usually fall into one of three. Why are companies making these if nobody wants them? What about privacy? And why do we

0:22 need external device if we have our phones? I ask all the same questions. [music] But one day I found this gadget in my post box. My default mode is being skeptical because I'm delusional enough.

0:34 I think that I can just build it myself. I did and I will show you how you can too. I'm going to share the hack how to record for hours and hours on Apple Watch which doesn't come by default.

0:45 Default is 10 minutes. Apple Watch goes to sleep and your recording shuts down. And because of me being skeptical and delusional, I just gave this memo gem to my fiance and I forgot about it. To my

0:56 surprise, turns out he's using it every day, loves it, and it was hard to take it away from him to make this video. So I asked him why does he love it so much. He clicked this one button and we

1:08 [music] just talked. Next day he just sends me a full transcript with ticky takeaways notes and this inspired me to build my own to address all my concerns and also take me as a sponsor but with

1:20 conditions which I was sure they are going to say no to. One, I get to investigate how this technology works. They have to answer all my questions. I'm going to build my own, teach people

1:30 how they can do it themselves. We are going to compare and they have no say into anything. I sent this brief and they agreed. They agreed. I click one button. I don't

1:42 know if you heard it, but it vibrated and there is a little red light indicating that it's recording. Right now on my watch, I click side corner and boom, it starts recording. I hope you

1:53 can see it. It's a little bit more indicative. We are going to be able to compare results at the end of the video. But I want to answer this question. Why? Why would you have a noteaker [music]

2:03 recording you or your notes? Obviously, this is just for me and I want you to understand how you can build it for yourself as well. I understand it's probably not for everybody and I fully

2:13 understand privacy concerns. That's why I want to control my data and wanted to build my own. Then I talked with my fiance like he works in sales in a bank. He can't have note takers in his

2:23 meetings. He can't even have note takers passively recording him in a call. But those calls have a lot of crucial details and sometimes those calls are back to back. So one use case he just

2:35 clicks a button after a call and just brain dams what he's thinking about. Another way that he's using he wants to write a book presses a button and speaks his mind. Another way our family

2:46 meetings we are planning our daughter's fifth birthday party. So, we just click record, discuss all the to-do items, planning, everything captured. We went on 7-hour drive to Denmark. He turned on

2:57 Memocat and I think it was capturing for 3 hours our whole discussion in a car. And when he was explaining to me why he loves it so much, we boiled it down to the ways that people like to think. For

3:09 example, for me, I really like writing. I submit my thoughts and I write. He doesn't like writing with his hand but he really likes to just thought to speech [music] right there is no penny

3:21 even in the middle. This is the most raw you can get your stream of consciousness and it's all captured. I also think about busy parents managing kids and they can just boom [music] press a

3:32 button your thoughts are captured. When he when he was telling me how much he loves it I'm like oh my god that's actually absolutely fantastic. I want that too. What I have on my phone is

3:42 also my AI assistant but it's living in Telegram. So the way I have to access to send a voice note I have to use my phone which takes your attention. I think that's that is one of the arguments why

3:53 [music] not phone. I think it's not so much about this device that device uh and I think everybody who wants it and want to control and own their thing you can do it now and I'm going to show you

4:06 how. The first direction that I wanted to go was actually phone because I thought that this is going to be way too complicated to do. I have iPhone 15 Pro. I think you can tap and it triggers a

4:20 shortcut. I had to give up that route pretty quickly because one of the shortcuts variables it just wouldn't save what I was putting in. So I was fighting it back and forward and then I

4:30 was just like, "Okay, I give up. Let's go for the device." I have my Apple Watch just laying around. This is pretty old one. I think it's series 9. You can just build an app on your Apple Watch.

4:40 And I'm going to walk you through how you can actually do and what you need to know. Now with AI coding assistance, that is not that hard. Um, you build an app, that app needs to connect to some

4:50 sort of service and database. And I already had my own memory system, my own AI assistant. So building an app and connecting everything, I was able to manage it under an hour. The important

5:01 caveat especially with Apple you have to have Apple developer account when you are building you have to enable developer and you have to download software called Xcode which allows you

5:13 to build apps for really any Apple uh device developer account yearly subscription is a 100 bucks you have to go through the whole application with Apple [music] that takes a little bit of

5:24 a time to get approved and then I would use clot codeex whatever you're using to build an actual app and then you connect your phone to that device. Enable developer mode both on your phone, both

5:36 on your Apple Watch. If you want to record for long time on your device, especially on Apple products. Apple categorizes apps into three buckets. One is default anything. Another one is

5:49 meditation apps which allow you to record in the background for up to an hour. The third one is sports is fitness app. So what I had to do if I want to just let recording go for multiple hours

6:03 and also when my Apple Watch goes to sleep enable in an app that it's actually a fitness app which then prompts me in a developer mode for the app to accept health settings. So that's

6:15 one workar around if you want it continuously record. The third route which I thought I'm going to take if first and second fails buying a cheap hardware device that cost around 14

6:26 bucks. I decided not to go with trying to build my own device. I think this was fun enough. Functionally, Apple Watch with my wrist noteaker is working. So, how does their gadget compare to my

6:39 watch? Apple when it records, it doesn't have all the bells and whistles to clean up audio. So, this audio capture from this device is way better than would be an Apple Watch or actually even your

6:52 phone. The Memo has dual noise reduction mics, so it can actually pick up audio 5 m from you. So that in their app, you can control the recording distance if it is close to you or even further to you

7:05 or it's [music] multiple people in a loud room. Uh on this Apple Watch has no noise reduction. What you could do, you can turn on Apple voice processing AV audio engine plus set voice processing

7:20 enabled. It is still the same mic. I didn't really bother to go deeper into it, but if you are in a noise environment with a lot of people speaking, yeah, Memo Cat is probably

7:29 winning that. So, this is 11 g. It's really light. Apple Watch is way heavier. I could feel it. This This one I almost don't feel it. And it comes with a strap that sticks like this. This

7:41 is not the best strap. [music] And I think it could be pretty easy to lose. Inside of it, it's a little thing. I'm wonder I'm now thinking if I can remove it. Wait, can I Is it still

7:53 record? Okay, so it's still recording and I'm sorry that it's nasty. It has been used. Okay, so here you go. This is a little thingy. The reason I got an idea with Apple Watch, they sent me

8:05 another strap. Okay, and this one you can connect to Apple Watch. I hope I'm not going to mess up the recording. I hope you can see. Both are still recording. Pop this one right in [music]

8:18 here. Okay. How does it look? It starts looking a little bit cyborgy. But because it's 11 g of a device, right, [music] you do not really feel like anything is more on your wrist. I think

8:30 the only factor is sleeves could be getting in a way. Both are still [music] recording. The argument for dedicated devices also pulls down to battery because now as it is recording, it's

8:43 draining my battery. If you're recording for long time, even an hour on your phone, that's going to grill your phone as well. Memoad battery, they say it lasts around 20 hours because it comes

8:55 with its own battery. It also comes with its own battery charger. So, battery lasts pretty much what they say. It definitely lasts a full day because you are not recording 24/7. But of course,

9:07 having another thing that you have to charge, that is one of the minuses. Where is Apple Watch, I already charged that one device and now it's no taker is inside of it. So that's something to

9:17 consider. But I don't really think that my Apple Watch even lasts full day or even last 12 hours. I'm actually not sure. And it's now recording for 22 minutes. Memo does ondevice buffer and

9:32 then syncs later either via Bluetooth or Wi-Fi. Okay, transcription. This is where we start touching the goodies of AI. I asked Memocadet how do they handle data and how do they handle

9:45 transcription? First thing for you to know the engine behind good transcription what Memocet is using is speech matics which by the way you can use it yourself as well. It's a company

9:56 and they provide this service plus they use assembly AI. These are specialized audio processing and transcription services that are on the cloud. These are another two providers that your data

10:08 actually travels to and they are processing it. My wrist note I went with something that I have available in my API keys and I'm using right now Gemini 3.8 flash [music]

10:19 and I'm routing it through one call. So very basic one call that does title transcription and then summaries as well on Google using Gemini if you could use API without paying for it but then that

10:33 data travels directly to the training nothing is for free if you pay and you opt out from the training they still processing and still having that data information. Another aspect is you can

10:44 use local processing whisper model. In in essence, if you would build your own app, you could also place a little voice processing model. There is very good models from Nvidia, especially designed

10:57 for devices. But again, that will drain your phone and drain your battery. So keep that in mind. Okay. So languages. This is so important for our household. I literally used wrist note to record

11:09 our conversation. It was Lithuanian, German, and English. and Gemini captured it pretty good. But Memoad is covering 153 languages. I think that's one other thing that sets apart from other

11:21 note-taking gadgets. And they even cover regional varants and native level accents. For my wrist note is Gemini and it's multilingual as well. My wrist note captures each language and presents it

11:34 Lithuanian, German, and English. And it doesn't translate. I have to translate that later. I would say this is where wrist note and I think both are transcribing pretty well. My camera card

11:47 ran out of space. Both recordings are at 58 minutes. I'm going to click stop on mine. So, because we just entered 1 hour, pause this and it's sending. And same on this device. It's still

12:01 recording. I'm going to press Okay, moment of truth. I literally just pressed Wait a second. You can see exact model how many audio tokens in and out was

12:14 used. It cost 8 cents and in total today notes that I recorded was 14 cents and it immediately presents me actions that I can take based on what it captured and detected. Here you go. Okay. So, how is

12:29 it on Memocat? Transferring recordings one out of four files. It is transferring right now through the Bluetooth. Does it mean that mine was faster? The thing is why mine is faster.

12:43 I'm not storing actual recordings. As you're going to see in Memo, it actually stores full audio file. You can always go back and check what was actually said and you can also run retranscription

12:57 with different models. So that is great. The first time when my parents are captured and he was like pressed played and I was listening I us talking in the background I was like oh this is weird

13:15 >> but I know that a lot of journalists use memo as a interview recording device because not only that it transcribes it actually has a raw audio especially when you do an interview you have to go back

13:28 and verify ify and check. In my case, I rely only on transcription. There is no way for me right now to go and verify and this is part of infrastructure to store actual recordings. You have to

13:40 store them somewhere. That takes space obviously. Then you have to process and that is where the latency is coming. So memo kit is GDPR compliant and we are now going through sock 2 compliance as

13:53 well. In my case because it's me there is no compliance. I don't have to give audit locks to anybody but again I [music] am exposing myself to security risk. I

14:05 have to cover all the cost. I have to protect my own data selfhost setup servers. I love doing those things but having somebody else handle and when things crash fix it is probably one way.

14:18 I think Memo is around 200 bucks for the device and then you have cloud subscription on the different tiers that you pay for with this all the audio services and AI features because those

14:31 tokens are pricey. In my wrist note, of course, I cover all absolutely all the costs from having self-hosted server, databases for free because I'm self-hosting, but I have monthly bill

14:45 from the server, all the API calls, transcriptions, additional tools and integrations, of course, all that I have to cover and it's unfortunately more than memo kit cloud subscription. But if

14:59 you don't have AI second brain or a system, I think memoad pretty good executed that they have memo chat and I really liked. Then I wanted to enable memo chat. Uh it actually told what they

15:12 using where data is going to be stored. It's AWS Asia using GPD models and then clock. Memo has contracts with cloud providers that we use that there is no training. You can delete your data at

15:26 any time using this memo chat. You can chat with all the transcriptions. It's kind of like a knowledge memory database. Memo actually provides a lot like a ton of integrations. Plus, if

15:38 they don't have baked in integrations like a notion, calendar, sana, email, they offer MCP and even CLI, which is like chef's kiss. The moment I see that a tool provides CLI and MCP that you can

15:53 literally connect to anything yourself if you know how to do it that is fantastic. So I love seeing that they have that capability. So really if you don't even have wristband and just use

16:04 it as a memory app you can integrate and connect that memory to anything like literally. Okay so we will put that to the side. Speaker handling this is super interesting. So if you are building

16:16 anything the transcription, meeting note taker, capturing notes, even voice notes capturing, you have to know this term diarization, I hope I pronounced it right. Identify who spoke and there is

16:29 multiple ways how to approach it. You can approach it just from prompting just instructing the model that hey based on the context think who spoke. You can also adopt a habit saying like I'm gonna

16:40 I'm speaking right now and kind of model can detect like back and forth. However, that is not obviously reliable. Memoad does auto speaker labeling. They are also cleaning the audio in its dedicated

16:51 device for the audio track. It's way easier to capture different speakers. The AI, however magic it is, it can't just guess who it is. It's going to identify speaker one, speaker [music]

17:03 two. The big problem happens when speaker one and speaker two identified correctly. But sometimes from speaker one the text the end of a sentence gets assigned to speaker B having named voice

17:18 identity says voice research uh they use that however that is biometric data so memo doesn't have enrolled voice print that it would detect based on voice quick update memo decided to do me dirty

17:32 so while I was editing this video memoad actually added voice print it's optin in and because it's biometric data, you have to confirm that you agree before you can use it. And by default, they

17:46 don't store it, which is honestly how it should be done. My voice note doesn't have any kind of detection. If you want to have enrolled voice print, piano pode AI and speech matics do that. So you

17:59 have to give around 30 seconds of a reference of a voice print like just a recording and just label that in [music] the future it should be able to detect that as well. It's pretty big

18:10 engineering part because then you have to also store that biometric data. If you are going to use any type of AI noteaker or device that actually remembers your voice, just keep in mind

18:21 they are storing biometric data and you probably want to check where and how your voice is getting stored or used. So um fixing wrong attribution in my setup I can't go and edit that. However, in

18:35 Memo they have an app and you can do that in in my solution. Anything above five different speakers start becoming one big mess. My intent with this, my Apple Watch is going to be just use it

18:48 for myself to capture quick thoughts. Instead of reaching for a phone, opening Telegram and sending voice note, I can now just press and record and then it goes directly to Telegram. So, that's

19:00 how I'm going to use it. But I absolutely love Memo app and that you can also capture notes through your phone transcription and they have so many features. So, we are going to now

19:12 just go through them. So here you go. Memocad building a personal AI noteaker. We have a brief and you can copy that as well. Uh we have summary that one pages is pretty cool. And then you can just

19:25 click share audio transcription brief everything and then decide when link expired. [music] If I go into transcript and I click anywhere it's downloading now the audio recording to my device for

19:38 the playback. Enable AI voice enhancements. Okay. enable and then you have all these additional templates, software requirements, specifications. So everything what I just said and

19:51 talked and hopefully I'm going to cut it short, it can turn it in specification documents. So >> rated and there is a little red light. If this mic failed or camera failed, at

20:04 least I have a backup. This is my go just also processed [music] the last note. Let's click this action. I hope it's going to go and find actual link of the company. So, okay. So, my go

20:17 executed. So, I did link to piano note and then boom, found exactly was open source project. Uh, I have a GitHub link and I have an actual link to their website. Okay. So, I'm going to link it

20:30 up as well. So, my plan is to use Memocat app, try to figure out to connect this built-in app, and then unless Memocat sends me another device, unfortunately, my definitely going to

20:42 steal Memo and just continue using. And I hope that this video either inspires you to check out Memoad or inspires you to get your hands dirty and build build a solution for yourself if you don't

20:55 like what other companies put together. Thank you so much for watching. See you in the next video.