Building Spatial Memory Part 4: Launching a Side Project — What I Learned After 3 Weeks of User Testing

Honestly, I didn't expect to be writing this fourth article so soon. When I started building Spatial Memory back in May, it was just a silly idea I had while hiking: "What if you could pin memories to physical locations, so only people actually there could unlock them?" Like a digital treasure hunt for your life.

Three weeks and four articles later (including this one), I've actually shipped it. I've got users testing it. I've learned more in these three weeks than I did in three months of coding alone. And I want to share everything — the good, the bad, and the "why doesn't GPS work in cities" ugly.

If you've ever thought about building a side project and actually shipping it, this one's for you.

The Idea Recap (In Case You Missed It)

For those new here, Spatial Memory is basically "Pinterest for the physical world." You pin photos and notes to GPS coordinates, and other people can only see them when they're physically within ~50 meters of that spot. It's like leaving little digital notes for the world to find when they show up.

Backend: Go + PostgreSQL + PostGIS + Redis GEO
iOS App: Swift + ARKit + CoreLocation
Storage: Cloudflare R2 (zero egress fees, hello side project budget)
GitHub: https://github.com/kevinten10/spatial-memory

I've already written about:

Today I'm talking about what happened when I actually asked people to use it. Spoiler: it's not all rainbows and "wow this is cool" tweets.

The Good: Some People Actually Get It

I sent TestFlight invites to about 20 friends and Twitter followers. About 8 of them actually installed it and created memories. That's a 40% conversion rate — way higher than I expected for a random side project with zero marketing.

What surprised me most? People got the idea instantly. I didn't have to explain it five times. One friend created a memory at a coffee shop we used to frequent in college, and left a note: "We spent 3 hours here debating AI and I still disagree with you." Another created a memory at a hidden hiking viewpoint with a photo of the sunset.

Honestly, that feeling when someone actually "gets" your weird little idea? It's better than any GitHub star. There's something magical about knowing your code enabled a real human connection that wouldn't have happened otherwise.

Here's what's working well so far:

Privacy design just works: Users upload photos directly to Cloudflare R2 from their phone. My backend never touches the image bytes. I don't store any personal info besides email (for TestFlight). Users love that — "oh, you don't get to see my photos?" Yes! That's the point.
Zero cost hosting: Everything's running on Fly.io with the free tier + Cloudflare R2. Total cost so far: $0.00. That's the beauty of using modern cloud services for side projects. You can actually ship something real without opening your wallet.
The AR experience is still cool: Even after building it, when you walk around and memories pop up on your screen in AR? It still feels like magic. People expect it to be janky, but when it works, their reaction is genuinely "wow."

The Bad: GPS Is Still Terrible In Cities

So here's the thing I learned the hard way: GPS works great when you're hiking in open mountains with clear sky. It works terribly when you're in a city surrounded by 50-story skyscrapers.

I kinew this going in. I wrote about it in the last article. But knowing it intellectually is different from watching three different users all say "why isn't this working where I am?"

One user was standing in Times Square trying to create a memory. GPS had him 80 meters away from his actual location. 80 meters! That's enough to put him in the middle of Broadway instead of the sidewalk he was actually on.

Another user was in a coffee shop in downtown Shanghai. GPS bounced between three different adjacent buildings. They tried five times to pin a memory and every time it landed in the wrong place.

So what did I do about it? Let me show you the fix I came up with that actually helped.

The Manual Override Hack That Saved My Launch

Instead of forcing users to rely purely on GPS, I added a simple manual adjustment mode. After the GPS places the pin, users can drag it around the map to correct the position themselves.

Here's what that looks like in Swift simplified:

// Before: Pure GPS automatic placement
let currentLocation = locationManager.location?.coordinate ?? defaultLocation
memoryPin.coordinate = currentLocation

// After: Give users control
struct AdjustablePinLocation {
    let gpsCoordinate: CLLocationCoordinate2D
    var adjustedCoordinate: CLLocationCoordinate2D?

    var finalCoordinate: CLLocationCoordinate2D {
        adjustedCoordinate ?? gpsCoordinate
    }
}

// In the UI: show a map draggable pin
let adjustableLocation = AdjustablePinLocation(
    gpsCoordinate: currentLocation.coordinate
)
// User drags to correct:
adjustableLocation = adjustableLocation.withAdjusted(newCoordinate)

It's such a simple change, but it completely changed the user experience in cities. Now when GPS is wrong (which it often is), users just fix it themselves. No more frustration. They understand GPS isn't perfect — just give them the tools to fix it.

Lesson learned: When you're relying on imperfect hardware/APIs, don't try to pretend it's perfect. Give users an easy way to fix it. Transparency beats clever algorithms every time.

The Ugly: Cold Start Is Still A Chicken-Egg Problem

Let's be real about this. Spatial Memory is useless if there are no memories where you are. If I open the app and nothing's around me, I'll close it and probably never open it again. That's just reality.

I have 8 active users who've created about 25 memories so far. All of them are in different cities around the world. That means 99% of the world has zero memories. So when a new user opens the app, they see nothing.

Is this a solvable problem? I honestly don't know for a side project. Let me think through what I've tried so far:

What I've Tried

I created memories in my own city: I walked around for an afternoon and dropped 15 memories at popular spots — parks, coffee shops, viewpoints. That helps when new users are in my city, but that's it.
I asked early users to create at least 3 memories: Most did, but they're spread out. One in New York, one in Vancouver, one in Singapore, three in Shanghai. Still very sparse.
I thought about seeding with popular locations: Like, automatically add "this is the Statue of Liberty" memory. But that feels cheap. The whole point is user-created memories from real people.

My Current Theory

Honestly, I think cold start is just something you accept with a side project like this. You don't need a million users. You just need a small group of people who actually care about the idea and enjoy using it. If it grows, it grows. If not, I still learned a ton building it.

So here's my current strategy: stop worrying about growth and just focus on making the experience better for the people already using it. Improve the GPS issues, add better AR placement, fix the bugs they report. If it's good, word will spread eventually. If not, that's okay too. It's a side project, not a startup.

Pros & Cons: Honest Assessment After Launch

I try to be upfront about this stuff — no marketing hype here. If you're thinking about building something similar, here's what you're getting into.

What's Going Well (Pros)

✅ Architecture choices have held up: Go backend is snappy, PostGIS+Redis GEO is still fast even with more memories, R2 direct uploads work flawlessly. I haven't had any production issues that weren't my own fault.

✅ Privacy design is a feature, not a chore: Users actually comment on how little data I collect. "You don't even ask for my name?" Nope. Just email for TestFlight. That's it. In an age of every app hoovering up your data, this feels like a breath of fresh air.

✅ Cost is zero: Still on free tiers everywhere. Perfect for a side project. You don't need to spend money to ship something real in 2026.

✅ The core idea still feels magical: When everything lines up — good GPS, good weather, a cool spot, and a memory from a friend — it's actually really special. I've had moments where I walked up to a spot my friend pinned, unlocked their memory, and felt that little jolt of connection that only this concept can create. That's why I built it. And it still delivers that.

What Still Needs Work (Cons)

❌ GPS accuracy kills the experience in cities: Even with manual adjustment, it's still frustrating. You start with the wrong position every time. ARKit world tracking is amazing, but it can't fix bad GPS data.

❌ Battery life is still bad: ARKit + GPS = you're lucky to get 2-3 hours of continuous use. If you're spending a day walking around exploring, you're gonna need a power bank. That's just the hardware we have right now, but it's still a poor user experience.

❌ Cold start is real: As I mentioned earlier, it's useless almost everywhere right now. That's not changing without a lot more users.

❌ App Store approval is still a hassle: I haven't submitted it yet. TestFlight is fine, but actually getting it on the App Store means dealing with all their review guidelines, privacy policies, etc. It's doable, it's just tedious. Side project time is precious — do I really want to spend a week fighting with App Store review?

What I'd Do Differently If I Started Over

Looking back at the past three weeks since I started this project (yes, I built the whole thing in three weeks — side projects don't need to take six months), here's what I'd change:

Ship earlier: I spent the first week doing "perfect" architecture. I should have shipped a rough prototype to testers in 3 days. You learn nothing from coding alone. You learn everything from users using it.
Add manual GPS correction from day one: I knew GPS was bad, but I thought I could "algorithm" my way out of it. I couldn't. Should have added this in the first version.
Don't overengineer the backend: I built a bunch of features I haven't used yet — like social feeds, friend lists, commenting. Nobody's asking for that yet. They just want to pin and discover memories. YAGNI applies to side projects too.
Start with a simpler non-AR mode first: I jumped straight to AR because that's what made the idea cool. But AR adds a ton of complexity. A simple map-first mode would have gotten me to users faster. AR can come later.

My Question For You

I'm still in the early stages of this experiment, and I'm genuinely curious what you think.

Have you ever used a location-based AR app like this? What was your experience? Did GPS drive you crazy too? Do you have any clever solutions to the GPS accuracy problem in cities that I haven't tried yet?

And more generally — do you think this idea has legs, or is it just a cool toy that'll never really take off? I'm still on the fence myself. I love the core idea, but the hardware limitations are real.

I'd love to hear your thoughts in the comments. Whether you've built something similar or just have an opinion, drop it below. I read every comment.

If you're interested in following along with the development, check out the project on GitHub: https://github.com/kevinten10/spatial-memory. Star it if you think the idea's cool — it helps more than you think.

Next up: I'm probably going to work on better battery optimization and a map-first non-AR mode for people who just want to browse without the AR magic. Let's see where this goes.

推荐订阅源

DEV Community