Dog whisperer

I have a running conversation with my dog, Max. On our morning walks, I’ll tell him about my plans for the day, my concerns, things I need to remember to do, etc. As it turns out, Max’s memory recall is even worse than mine, so I started recording voice memos with the Voice Recorder app on my iPhone. When I’d get home, I’d copy the transcription and paste it into ChatGPT: “Please clean up this audio transcription, and create a list of key topics and to-dos.” It technically worked, but it was clumsy and required too many steps. I soon ended up with a backlog of dozens of unprocessed voice memos on my phone and gave up.

That experience gave me an idea to write an app which would automate the process - an AI audio diary. I started playing around with an idea: I could use a more robust audio transcription service, and I could hook in to my phone’s APIs to create reminders, calendar events, etc. The more I dug in, more questions started popping up and I started having second thoughts. The space seems quite crowded, and sending your most personal thoughts to some SaaS provider who in turn ships them off to OpenAI… I don’t know. Seems a bit sketchy. I’m certain that there’s a way to do it properly, but it seemed to me that I’d end up spending a lot of time trying to convince users that I wasn’t collecting a bunch of potential blackmail material. I still might end up doing it for myself. Who knows?

These days, I’m mostly a backend developer. Occasionally I’ll sub in on the front end, but I’ve never worked on a mobile app before. Not written a single line of code in Swift. For the voice recorder, I was interested in exploring the performance benefits of native mobile development, so I spent some time looking into the various iOS and Android APIs. I was especially intrigued by the amount of data available through Apple’s HealthKit. In comparison to Google’s Health Connect, it’s a treasure trove of information which can be used to offer insights to users (with their permission, of course!).

A pivot

An idea began to take root. Instead of processing a diary entry (“I feel really worn out from my workout yesterday, I don’t think I got enough water”), I could record commands, and using an LLM as a router, determine the code to be executed on the backend. I could save those changes on my server, and then push the change back to the client and up to HealthKit. Using voice as the primary UI, I could write code that could record any of the metrics provided by HealthKit. There are hundreds (thousands?) of health and wellness apps on the App Store, but they all stay in their own lane: hydration, diet, mood, etc. If my app brought all that together, could it work in concert, the whole being more than the sum of its parts? I could add my own data types as well! While they wouldn’t be synced up to HealthKit, you couldn’t view them in the Health app on your phone, but you could still use it to provide more insight to the user.

Here’s a good example. I’m a middle-aged guy, and while I like to think I’m aging gracefully, I just cannot perform physically like I used to. I’m still strong, but I’ve noticed that muscle fatigue has become more and more of a “thing.” If I go too hard at the gym, I will feel the fatigue for days. But what if the problem was that I actually just wasn’t hydrated enough? What if I need more protein? More sleep? Similarly, I’ve noticed how dramatically lack of sleep can impact my mood and focus. One sleepless night can wreck my mental state for… I’m not exactly sure how long? Maybe the data could tell me.

The way forward

So that, there, is the real kernel of the idea. What if I could make it so that I could make it as easy as possible to record as much about my daily habits that I could have enough data to provide answers to those questions? What if, after logging a sleepless night, my app could tell me that, according to the data on hand, I can expect a lower mood and difficulty focusing for the next 36 hours? Additionally, the data shows that light exercise, sunlight exposure of greater than 90 minutes, proper hydration, and meditation has historically improved my overall wellbeing. The app could recommend a course of action to counter the forecasted low mood.

That’s it - that’s the idea. We’ll see if it works. In the very least, I’m excited to try find out.