GadgetsTechnology

Open HN: CamelQA (YC W24) – AI that exams mobile apps

Hiya HN! We’re camelQA (https://camelqa.com/). We’re building an AI agent that might perhaps perhaps perhaps automate mobile devices the utilization of computer imaginative and prescient. Our first use case is for mobile app QA. We convert pure language test situations into exams that bolt on accurate iOS and Android devices in our instrument farm.

Flaky UI exams suck. We are desirous to uncover a resolution where engineers don’t destroy time affirming fragile scripts.

camelQA uses a combination of accessibility part recordsdata along with an in-condominium custom imaginative and prescient-simplest RCNN object detection mannequin paired with Google Siglip for UI part classification (deem a pair of sample output right here – https://camelqa.com/weblog/sole-ui-part-detector.png). This skill we’re able to detect aspects even if they uncover now now not get grasp of accessibility aspects connected to them.

Under the hood the agent is the utilization of Appium to interface with the instrument. We use GPT-4V to motive at a high stage and GPT-3.5 to attain the high-stage actions. Verify out a gif of our playground right here (https://camelqa.com/weblog/sole-signup.gif)

Since we’re imaginative and prescient essentially based completely mostly, we don’t need uncover entry to to your provide code and we work all the draw through all app kinds – SwiftUI and UIKit, React Native, Flutter.

We constructed a demo for HN where it is possible you’ll perhaps use our mannequin to govern Wikipedia on a simulated iPhone. Verify that out right here (https://demo.camelqa.com/). Are trying giving it a job love “Bookmark the wiki web order for Ilya Sutskever“ or “Opt up San Francisco in the Locations tab”. We simplest get grasp of 5 simulators running so there can also probably be a wait. You uncover 5 minutes whenever you enter your first issue.

In issue so that you just can leer what our front cease looks to be like love, we made an memoir with some test runs. Exhaust this login (Username: hackerNews Password: 1337hackerNews!) to behold our sandboxed HN memoir (https://flee.camelqa.com/login).

Final year we left our company jobs to produce in the AI condominium. It felt love we had been with out waste testing our apps, even for minor updates, and we aloof shipped a worm that introduced about regarded as one of our apps to smash on subscribe (the app in ask – https://apps.apple.com/us/app/tldr-ai-summarizer/id644930471…). That modified into as soon as the catalyst for camelQA.

We’re furious to listen to what you all deem!

Read Extra