cheetah - AI-powered Mac app for enhancing remote coding interviews

Cheetah: Real-Time Interview Assistance for Software Engineers

Cheetah is an innovative macOS application specifically designed to assist users during remote software engineering interviews. It offers real-time discreet coaching and integrates seamlessly with live coding platforms, making it an excellent tool for candidates looking to enhance their interview performance.

Overview

Cheetah aims to help users improve their chances of securing high-paying software engineering roles ($300k SWE roles) by providing support during interviews. This support eliminates the need for intensive weekend cramming of algorithm challenges, often found on platforms like LeetCode, which some users find overly demanding.

How It Works

The application operates by leveraging two powerful technologies:

Whisper: A tool for real-time audio transcription. This feature ensures that all details of the ongoing interview are captured in text format instantly, allowing users to review and analyze the conversation effectively.
GPT-4: Offers intelligent hints and solutions based on the transcriptions. Users must have their own OpenAI API key to access GPT-4 features. In cases where GPT-4 is unavailable, an alternative version, gpt-3.5-turbo, can be utilized.

To function at its best, Whisper requires a recent M1 or M2 Mac, as performance is optimized on these devices. Whisper is implemented via Georgi Gerganov's open-source project, whisper.cpp.

Getting Started

Prerequisites:

macOS version 13.1 or newer is required.
Installation of whisper.cpp, and the SDL2 library via Homebrew:
```
brew install sdl2
```

Audio Driver Setup:

For optimal performance, it's important to ensure that your audio input captures both ends of the conversation. Using video chat applications like Zoom or Google Meet? You can use BlackHole, a free audio loopback driver, to set up a multi-output device, ensuring the application functions smoothly without using the loopback device as an input for the chat app itself.

Application Features

Upon launching Cheetah, users can:

Select an audio input for live transcription.
Make use of the app's user-friendly interface featuring three main buttons:
- Answer: Generates potential answers for the interviewer's questions.
- Refine: Updates the answers, particularly useful if the interviewer adds more constraints.
- Analyze: Evaluates the code and logs within your live coding environment, requiring an extension on your web browser.

Additional functionality includes the ability to select part of a generated answer to refine it further for more detailed insights.

Installing the Browser Extension

Currently, Cheetah supports Firefox through a specific browser extension. To install it:

Modify the matches field in ./extension/manifest.json with the domain of your live coding platform.
Open Firefox's about:debugging.
Select "This Firefox."
Load the extension temporarily by selecting ./extension/manifest.json.

Disclaimer

It's important to note that Cheetah is a satirical art project and is not intended for real-world use. The application may produce incorrect or inappropriate outcomes, and users must exercise caution and take responsibility for the app's outputs.