Hacking the Coding Interview

Many engineers think that coding interviews are useless!
Image Based UI
Text Based UI
Output answer to a coding question typed in
Output answer to an uploaded coding question
Output answer to an uploaded coding question
Audio Based UI
Output answer to a coding problem discussed during an interview

Inspiration

We, like most other software engineers, were frustrated with the unnecessary nature of coding interviews and how irrelevant they were to real-world software engineering ability.

What it does

This app can sit in the background on another device as you log on to your video interview. Press record, and it will output a summary of a solution, as well as python code for the problem.

There are three functionalities at the moment:

Live help during a video interview. Record the interviewer's question and clarifying questions you have, and it will output the solution. You can also record later parts of the interview, such as follow-up questions or curveball questions, and the app will adapt to them.
Upload a screenshot of a problem and get the solution. Doing a take-home screener that disabled copy-paste? Use this feature!
Enter a question by text and get the solution.

How we built it

We built a sophisticated prompt pipeline to construct the solution "summary" as well as the python code to the problem. We did not want to just output code as a solution, because that is impractical to read in a high time-pressure environment like a video interview. Instead, we output solutions in natural-language, and bold key elements of the technique so that you can quickly the deduce the answer from just a single glance.

Otherwise, tech stack is Python, FastAPI, and Gradio.

Challenges we ran into

The UX of this app was a challenge that we had to grapple with. How can we present information in a way that the interviewee can process as quickly as possible under the stressful environment of a video interview? We settled on outputting natural-language solutions, and emphasizing important parts of the solution in the UI.

Moreover, building the prompt pipeline to be reliable was also a challenge.