I Built a Minimalist AI Note Device

0:00 This device here is basically my new second brain because I constantly have ideas that disappear again 5 seconds later. And maybe you know this problem too. You think of something important, maybe an idea for a project, something you need to do later, or just a random thought you don't want to forget. But if you don't write it down immediately, it's gone. And the problem is your phone is honestly terrible for this. Every single time I unlock my phone to quickly note something down, I get distracted.

0:28 So, I started building this. And when I first thought about this project, I realized that the requirements for something like this are actually pretty simple. You basically need a microcontroller, a display, a microphone, a speaker, a battery, some kind of storage, and decent power management. Hm. Okay, that's maybe a little more than simple, but then I came across this board from Wave Share, and it turned out to be pretty much perfect for the project. For anyone who doesn't know, the ESP32 on this board is basically the brain of the entire device. It's a super popular low power microcontroller that's perfect for DIY projects like this. It has built-in Wi-Fi, Bluetooth, and enough performance to handle all the logic for the device itself.

1:15 And then there's the e- in display. What makes e- in so perfect for a project like this is that it only really consumes power while updating the screen. So once something is displayed, it can basically stay there without constantly draining the battery like a normal display would. It also just feels much calmer and less distracting to look at compared to a bright phone screen. And the nice thing is that because of this setup, the project is actually really approachable to build yourself. I designed the entire thing around parts that are easy to get and simple to assemble. You can basically build the exact same device yourself without needing a ton of experience. All of the files, firmware, and assembly instructions are available on my Kofi.

2:00 The first thing I worked on was the case. The board already comes with its own enclosure, but I really didn't like the shape of it. It felt too boxy, and you could only use very specific battery sizes. And after the tiny e-reader project, I also noticed that people really enjoy customizing and modifying these kinds of devices. So, this time I wanted to make something that's easy to print, easy to customize, and requires as few additional parts as possible, and most importantly, no screws. I went through a lot of different prototypes because I wanted the whole thing to snap together nicely while still feeling solid enough to throw into a backpack or pocket every day. And after a lot of testing, I eventually ended up with this design. At first, it looks really simple, but there are a few details I'm really happy with. The two case halves snap directly into each other using these little clips on the side, so the entire device locks together without screws. Inside, four small pins hold the board securely in place, and the buttons simply drop into these cutouts here. I also rounded basically every edge and corner because I wanted the device to feel soft and comfortable in the hand.

3:34 For the battery, I used a small 500 mia lipo because I wanted to keep the whole device as compact as possible. But technically, you could also use much larger batteries if you slightly increase the case size. And one thing I already learned from the tiny e-reader project is that these kinds of devices usually evolve way beyond the original version once the community gets involved. I actually had to delay this video multiple times because there were still bugs and weird issues I wanted to fix first. Especially the audio recording and power management gave me way more problems than I expected. I originally thought this would take maybe a few evenings. It definitely did not. But now I'm finally really happy with how it works. So, let me quickly show you the interface. This here is basically the deep sleep screen. The device spends most of its time in this ultra low power mode to save battery. And once I hold the button, it wakes up and brings me to the home screen. The ring around the display shows the battery level. And from here, the interaction is intentionally extremely simple. If I hold the upper button, the device starts recording. Important note, don't start another side project. As long as I keep holding it, it records everything I say.

4:50 And once I let go, it automatically saves the recording, gives a little sound confirmation, and opens the tag menu. So, for example, if this is an idea for a future project, I can simply select the project tag and after that, the device immediately goes back into deep sleep again to save battery. So, the whole interaction only takes a few seconds. At this point, we basically built a really tiny voice recorder. And honestly, even this version already worked surprisingly well. Everything gets stored directly onto the SD card.

5:19 You can listen to recordings again because the device also has a built-in speaker. Important note, don't start another side project. And overall, the experience already felt really polished. But there was still one major thing missing for me, transcription. Because voice recordings are great for quickly capturing ideas. But once you have a lot of them, searching through audio files becomes really annoying. And since this entire project runs on an ESP32, there was no realistic way of doing the transcription directly on the device itself. So I had to find a different solution. And this is where the AI integration comes in. After trying a few different options, I eventually decided to use OpenAI Whisper for the transcription. The setup itself is actually really simple. Before flashing the firmware, you just enter your Wi-Fi credentials and OpenAI API key into the config section of the code. And once the device connects to your home Wi-Fi, it can automatically upload the recordings and transcribe them. Test test test. To save battery, I specifically didn't want the device constantly using Wi-Fi in the background. So, right now, it either synchronizes automatically once per day or you can manually start the sync process from the menu. And this is what that process looks like. The device uploads all recordings one after another. Whisper transcribes them and then the text files get stored directly back onto the SD card. So now every note exists both as an audio recording and as readable text. And the really exciting thing about this is that it opens the door for way more features later on because now the device already has an AI connection built into it. So future versions could automatically summarize notes, generate titles, categorize ideas, or maybe even create little daily overviews. But there was still one more thing I wanted to add because now we have voice recordings and transcribe text directly on the device itself. But what if you actually want to use those nodes somewhere else? Maybe inside notion, Obsidian or your own workflow.

7:23 So for that, I also created a transfer mode. Once activated, the device connects to your home network and displays a local IP address on the screen. You simply type that IP address into your browser and this little web interface opens. And I tried to keep this part just as simple and minimal as the device itself. You can browse through all your notes, filter them by tags, download the text files, or export the audio recordings directly from the browser. I also added a few smaller quality of life features into the firmware. For example, optional button sounds.

8:04 At first, I thought this was unnecessary, but especially because e- in displays are relatively slow, the sounds make the interaction feel much more responsive and tactile. It's just this small bit of feedback that makes the whole device feel more alive. And yeah, that's basically the PA Note. So, let me know what features you would add next. And I'm also really curious what you think about this idea. Would it make sense to bring parts of the tiny e-reader firmware over to the Pala node as well? Or do you think the display is just too small for that? Because now that people already started building apps for the e-reader, I feel like there's a lot of potential here, too.

8:43 Anyway, thank you so much for watching and I'll see you in the next video.