Olpaka 0.5: The Scientist

September 28, 2024 5 minute read

Nobody said it was easy
Oh, it’s such a shame for us to part
Nobody said it was easy
No one ever said it would be so hard
I’m going back to the start

It’s been a while since I’ve published something huh!?

Well, as per premise in previous blog posts, I’ve lost interest into this project. So I’ve decided it was better to just leave it there for a while.

Up until a friend of mine told me that Kotlin Multiplatform + Compose work like a charm. Something similar happened at work. I had to give it a try!

Since I cannot be bothered to clone twice the same project, I’ve spent the past month rewriting the whole project in Kotlin Multiplatform. So, some elements might look the same as before, others slightly different, new bugs might be there … ENJOY!

What’s new

Connection settings

Well, from a user experience, there should be pretty much nothing new.

What you’ll find different is just a new section under the Settings tab where you can connect to any ollama endpoint.

For instance, I have an instance deployed on my Raspberry PI and that’s where I use it with some insignificant models such as Qwen2.5:0.5b. At the end of the day, I had to add at least one functionality to justify a new release. Right!?

The discord server is now working!

Earlier I already had a Discord server link on the Readme.md file. However, I didn’t notice I’ve put an expiring link. That’s why the server was so lonely!

If you want to join me, have a chat, report a bug, argue about why the sky is blue, join the official Olpaka Discord server!

How is Olpaka doing?

Well, it’s definitely better that expected, considering there was a critical bug that would prevent people with no models to download their first model! In the past 3 months Olpaka has been used:

🌍 by 300 unique users from 57 different countries;
💬 to send 100 messages - mainly using llama3:latest;
⬇️ to download 30 models - mainly llama3.1:8b\

(NOTE: I’ve used emojis, so it looks more appealing to the audience: I’m finally doing some progress in self-marketing as well 🥳 ).

Hopefully, with the new version things will be better.

Nerd stuff - Flutter vs Kotlin Multiplatform

Premise

I’ve been using Kotlin pretty much every single day for the past 8 years, so it’s obvious I’m more fluent with Kotlin therefore I’m biased.

The technology

Flutter is way more mature than Kotlin Multiplatform. This is a fact: Flutter has now been around now for about 7 years, therefore it comes with tons of multiplatform libraries that just work out of the box, plenty of blog posts, StackOverflow questions and documentation.

Despite I don’t really code with AI on a daily basis I’ve found ChatGPT really struggling with Kotlin Multiplatform - especially WASM as it’s an experimental technology.

So, all I’ve got was wrong answers, 2 pages of documentation from JetBrains and a Slack channel to ask obvious questions to which the answer was either you can't do it or Yes, build it on your own.

Olpaka is a very trivial piece of software as it’s mostly:

Render some UI;
Render some Markdown;
Dependency Injection - nice to have;
Performing network calls;
Storing few key/value pairs;
Collecting some anonymous analytics;

So, let’s make a little comparison

Feature	Flutter	Kotlin MP
UI	Material	Compose
Markdown	markdown_widget	markdown-renderer
Dependency Injection	get_it	koin *
Network	http, dio	ktor
Preferences	shared_preferences	multiplatform_settings
Analytics	firebase	firebase / custom **

* Spent 6 hours fiddling with it as it prevented me to run it on WASM when the app was not deployed on localhost 🥔

** There’s no such thing as Firebase Analytics multiplatform. So I had to import it on web, then create the bindings JS/Kotlin and provide an implementation. For desktop I’ve just implemented the actual Google Analytics reporting protocol. Both of those approaches are quite painful.

So, as you can see, Flutter already provides out-of-the box most of the tools you might need to get started.

Dart vs Kotlin

Here is where things become obvious. Let’s just face it: if Kotlin is a sugar-coated version of Java, Dart is definitely the evil twin of Java.

Weak type system

The Dart’s type system is very weak in general as I’ve encountered a huge number of type casting nonsense - sorry: I can’t remember exactly what was the scenario.

Compared to Kotlin, there are no sealed classes - yes I know you can download a library that mimics sealed classes, however, mimicking is not exactly the same as having the concept built into the compiler!

It wasn’t as terrible as Python or Javascript, but for sure suboptimal.

Code generation

With kotlinx serialization you just need to annotate your data classes, and you’re pretty much good to go! With Dart, you’ll need to fiddle so much that I gave up and parsed the JSON manually.

Maybe it’s just me and my lack of patience here.

Resource usage and compile times

Well, when it comes to Kotlin compile times are on another level.

I’ve got a pretty decent Intel 8th gen computer with 32GB of RAM and SSD and the kotlin compilation was painfully slow. A fresh web compilation takes around 5-8 minutes, a fresh desktop takes about 2-3 minutes.

Subsequent builds are way quicker, however, are not as instant as it happens with Dart.

This is definitely a score from Dart.

UI Preview

When you are new to a UI Framework you might want to have a quick visual preview of what you’re building. Well, it might come to a surprise, but none of the two technologies give you a decent live preview of what you’re building!

That said, at least Dart has an “instant refresh” triggered when you press CTRL + S, and it works on web, desktop and mobile. Kotlin multiplatform provides such functionality only when the project is opened using the Intellij’s Fleet IDE. The only con of this tool is that it’s not just slow on its own, but it will also slow down or freeze your entire OS.

Actually, there is a rather surprising workaround to this issue.

Other perks

I’ve pretty much missed the concept of data classes, especially the copy utility function. Again, I know that there are libraries to mimic this, but it’s like having a carbonara with bacon (guess where I’m from)!

async/await can be error-prone because I often forget to add the await keyword.

However, one thing that I really loved about Dart is the lack of an obvious multithreading mechanism. If you want to achieve true multithreading, you’ll need to put some serious effort. This is definitely a PRO as most of the developers do not think (know?) about concurrency when developing an app.