Kyle Wascher

Kyle Wascher

@waschlogic.com

Current: #ReactNative #ComputerVision #ML #typescript #android #ios Past: 14 years of corporate dev, #csharp #dotnetcore #react #sql #MLS #ColumbusCrew #USWNT #USMNT #Fishing

179 followers747 following130 posts

Anyone up for some Michigan fishing? Hoping to connect with fellow anglers for feedback on a mobile app I've been working on. Casual hangout, beautiful scenery, and hopefully some fish! Let me know if you're interested!

FishBytes.app@fishbytes.app

With all he rain we've had in Michigan, the rivers are blown out. So I explored a new (to me) lake yesterday! Managed to locate a few largemouth to keep the skunk away! #michiganfishing #kayakfishing

Looking for more anglers (on android) to help test the new features! My goal is to make the entering a catch as seamless as possible. Its subtle here, but after the picture is taken on device it runs some ML models to detect the fish species and detect if a measuring board is in the photo.

FishBytes.app@fishbytes.app

Made it out on the water for the first time this year! I've been working on the catch a fish -> take a picture -> measure flow. Pretty happy where its at, just a few bugs when the screen rotates to cleanup!

If there is a measuring board then it zooms in to where the fish tail is to get a better view, and allow you to quickly enter the length. If the species detection is poor or wrong, another screen is available to enter it.

Just came across github.com/javierbrea/e... yesterday. I'm sure its overkill for some, but allowing you to setup custom rules for what layers of code can import is really nice for me. Example: I can ensure that my react components do not directly reference my database layer.

github.comGitHub - javierbrea/eslint-plugin-boundaries: Eslint plugin checking architecture boundaries between elementsEslint plugin checking architecture boundaries between elements - javierbrea/eslint-plugin-boundaries

Very sold proof pre-training dataset matters per task.

TimDarcet@timdarcet.bsky.social

We test 2 dataset settings: “pretrain on ImageNet1k”, and “pretrain on bigger datasets” In both we significantly improve over previous models Training on a Places205, is better for scenes (P205, SUN) but worse for object-centric

Kyle Wascher reposted

Throwback to my Master Angler patches from 2021 and 2022. It's been two years, but every trip on the water is an adventure, and I'm having much fun chasing the next one! Tight lines, and maybe a new patch in 2025! #michigan #masterangler #fishing

One of the best things about TypeScript? Turning business logic into compile-time errors. The `never` type is a game-changer for catching issues early. How are you using it in your code?

I bought a bottom tier M1 Pro in 2021 to see for myself if this new ARM thing was going to work. I also had just became dad for the first time and portability was also important. Just replaced that M1 with a M4 max. 🔥 Huge upgrade.

This past year I 100% switched from windows to Mac. My windows desktop is a now running Ubuntu and its main use it training deep learning networks. (With a side of gaming on Friday nights)

Year one review of daily development with #ReactNative. 1.✅ Yes use Expo CNG 2.✅ Don't over complicate things now for later features. 3.🤔 Add libraries/dependencies sparingly. Ecosystem moves fast and things break. Or if your desired behaviors don't match. (Ex. Wrote my own fully typed router)

View full thread

5.❌ You will lose countless hours to xcode & Android studio updates/bugs. (Just part of the software dev lifecycle.. had similar qualms o seen the microsoft stack) 5.5😵‍💫 lost time to app store red tape 6.✅ The speed of creating new screens and live hot loading while building is 🔥

+1 🦀 I recently started using Rust's Result<> style pattern more in TS and really liking it so far. Having all error states defined is a game changer. (Only issue is finding all those states in a try/catch world)

Year one review of daily development with #ReactNative. 1.✅ Yes use Expo CNG 2.✅ Don't over complicate things now for later features. 3.🤔 Add libraries/dependencies sparingly. Ecosystem moves fast and things break. Or if your desired behaviors don't match. (Ex. Wrote my own fully typed router)

4.❌ TS -> Native Code & Native Code -> TS interfaces are still manually managed. Turbomodules generates methods... but you still have to manually pull off properties and type them etc. (I ended up writing my own codegen script to type these fully. along with events. No future bugs from typos thx)

You are up to date.