Supio Software Engineer Interview Experience — One Hour Live-Debugging Their AI Chat Codebase

A personal injury law AI startup in Seattle.

One hour of debugging. The interviewer emails you a source code repository, you run it locally yourself, and fix the issues. You need to be reasonably familiar with how to work with some AI APIs. It's a chat interface.

• Bug 1: the console showed "API key invalid." The cause: a wrong API key.
• Bug 2: after sending a request in the chat box, the response that came back rendered empty. The cause: the returned text was in the tokens property, but the local code was reading the text property.
• Feature 1: add a clear button — clicking it clears the message history and context and reinitializes. There was already an action in the store I could just call.
• Feature 2: add a stop button that terminates the stream while OpenAI's response is still streaming in, similar to the stop button (the little square icon) in ChatGPT. This one was about state management.

Even though this kind of interview is closer to actual work, since I wasn't familiar with the tech stack it still took some time to adjust and understand things, and a case like bug 2 (where the type of data the API returns doesn't match what you locally expect) really did require shifting my thinking before I could see what was wrong.

There were probably more feature requests after that, but I ran out of time.

Discussion part:

The company is expanding its hiring — they recently brought on some UI designers (huh, wasn't this all supposed to be replaced by AI, ha). They're short on developers, and the on-call load is heavy. The development cycle process is just starting to get established, to avoid the old situation where the PM would directly force urgent features onto the team.