Thoughts on a new Cycling App I'm developing?

At first I was negative about this, especially since I think “AI” is incredibly over hyped.

But reading about how much you have / will customize it for cycling and all the possible use cases, I think it’s actually pretty neat.

Here’s the things that I would want assistance with that are new capabilities that I don’t think anything else does:

  • Grab my power-duration curve from intervals.icu (or similar) and compare it to what I’ve done on the current ride to give me some metric for how fatigued I probably am.

  • Using the above and a known route and your physics model, suggest if I should ease up to be able to finish without bonking

  • Something like Best Bike Split power plan except have it adapt in real time based on how I’ve ridden so far and give me suggestions every 10 minutes to go a bit faster or slower (basically similar to the above, but proactively trying to optimize for best speed my fitness allows).

  • it’s a big stretch, but if I give it inputs (on screen) before or after the ride on the clothing I used, and give feedback on how suitable it was, correlate that with weather and ride type and expected speeds to remind me with that data for what to wear on future rides in similar weather. Example: I wore a heavy thermal jacket and thermal tights on a 50F day on my MTB and said that was too hot. But I wore the same on my road bike on a different day and it was perfect for the same weather. Today is 45F MTB day it will remind me what I write and how I felt on that 50F day so I can figure out what I should wear.

Making actual suggestions for what to wear would be the next step, but that’s probably way, way harder to do. Simply surfacing relevant past data would be immensely helpful.

2 Likes

First of all, I’m glad you came around! I too am really starting to hate AI integrations where there doesn’t need to be one, AI is essentially the new crypto at this point.

However, it does have some magic when used for only things it can do, not just poorly summarizing websites or creating spammy garbage.

An example being, I just added another AI approach for realtime audio denoising, it decimates non AI algorithms. I can stand directly next to my air compressor turning on (extremely loud) and it eliminates all background noise.

Moving on. You’ve nailed some of the possibilities of such an app, not only can it give you splits for upcoming hills/segments, but it could monitor, figure out you’re in a pack that’s too slow for you and mention at what mile an upcoming breakaway opportunity may be present.

Beyond this, I LOVE your idea for it taking down notes as to what you feel in what condition weather wise with what clothing. It’s very hard to get it right and I’m far too lazy to whip out a note pad and aggregate results. I’ll likely find some way to integrate that.

1 Like

That’s why I much prefer “machine learning” terminology over AI. I believe ML is the more accurate term for things like that.

As far as integrating with any cycling data platform, I recommend intervals.icu since it’s free, already connects to many other platforms, is popular, and is run by a single developer that’s very responsive to user feedback.

My humble suggestion would be to make an Android app first since you don’t have a Mac. Prove it out and figure out your UI on Android. At that point you could ask for donations to help fund the Mac purchase. (Said as someone with an iPhone. Typically I would say iPhone first if you were trying to be profitable).

I’m planning on looking into various integrations, I already have an integration with Garmin for route-specific API, but am looking into similar mechanics with Strava/Wahoo/Garmin for activities. Additionally, it would be trivial to do a lot of the calculations that a site like Intervals.icu does, the real challenge of a site like that is meaningfully displaying the data with a complex, interactive, data driven UI. However, this app, at first at least, won’t have much of a UI and will be designed to stay in the pocket, so it could access any of those data sources and provide rich feedback while cycling, accounting for your current fitness/fatigue/etc.

Further, yep, I’ve already built out a proof of concept on Android. I had to make sure I could get these ML models working on the phone, which was quite challenging, but doable.

The idea is, build a bare bones, functioning Android one that I could use and mess around with, or, spend the extra time time to build it into a solid production app and then an accompanying IOS app (perhaps even an Apple Watch app) if this idea had any validation.

Also, personally, I don’t see a distinction between “AI” and “ML”, I think the appearance of large language models makes it seem like there is a distinction, but in the end it’s all using tensors for input/ouput, attention mechanisms, convolutional neural network layers, etc. Interestingly, I use CVT, a convolutional vision transformer, which, internally, is very similar to a LLM like ChatGPT to determine road surface type from satellite images for my main routing site. This works alongside a Resenet152 model, which is far more common for image classification, similar input, and similar output, but by your definition, the first would probably be considered AI, and the second ML.

No argument here by the way, I just find it a fascinating conversation, and I do agree there should be some distinction between chat bots and something like image classification.

IMO I wouldn’t bother with any work regarding displaying and helping the rider analyze data. TP, Garmin, Strava, Intervals.icu all do that well and few people are likely to care your app can too.

However, if you want to know what the user is capable of (FTP, PD curve), then you will need to integrate with something that has that data. That’s the only context that I mention the intervals.icu integration.

I agree the interface is the app itself should be pretty sparse, but you’ll at least need some text to remind / inform people the app’s capabilities because they (or at least I) will have trouble remembering what I can ask it do.

I totally agree, I also hate making GUIs (although, I don’t, I have a brother who does all of that for my various sites).

Also, I was envisioning, if I did make a GUI, it should show your route and have graph capabilities so you could state “show my my power curve”, and it opens that menu or “where’s the biggest hill in the course” and the map zooms into it, pulls up a graph, and gives details.

But again, I doubt I’d go that route, I might, but it’s better to focus on one thing, like the best voice communication/answer app, then try to be an everything app.

1 Like

This sounds really interesting to me. I currently use my iPhone in place of a head unit and have struggled with the value proposition of a dedicated head unit. If your app can log ride data (gps, watts, and hr), it would be perfect.

Are you still working on this? I’d love something like this. I don’t know a lot of other female road cyclists so I’m often riding solo. And I have a poor sense of direction so I’ve gotten lost many times - which is embarrassing but also can be scary as it gets darker and you get more tired. Trying to figure things out on a little map is frustrating as well. I just use my phone so I’d love to not need it taking up space on cockpit.