RubberDuck2.0

Watch the demo on youtube!

Inspiration

Preparing for technical interviews can be intimidating, especially for students who don’t have access to frequent mock interviews or detailed feedback. Many candidates know the material but struggle with communication, confidence, and structuring their answers under pressure. We were inspired to create RubberDuck2.0 as a tool that lets students practice interviews in a safe environment, reflect on their performance, and improve both their technical and behavioral skills before a real interview.

What it does

RubberDuck2.0 is an AI-powered interview practice platform designed for software engineering and computer science students.

The platform allows users to:

Practice technical and behavioral interview questions
Record spoken answers, which are then transcribed to text
Receive AI-generated feedback and scoring on their responses
Solve LeetCode-style coding questions directly on the platform
Get vision-based coaching insights such as presence, smile, and professionalism

By combining spoken explanations, written code, and feedback, RubberDuck2.0 simulates a realistic interview workflow.

How we built it

We built RubberDuck2.0 as a web application using:

React and TypeScript for the frontend
Vite for fast development and hot reloading
Browser APIs (MediaRecorder, getUserMedia) for audio recording and camera access
Speech-to-text to transcribe recorded answers
Large Language Models (LLMs) to generate interview questions, evaluate responses, and provide feedback
An integrated coding editor to support LeetCode-style technical interviews
Lightweight computer vision metrics to provide real-time behavioral coaching tips

Challenges we ran into

Achieving reliable speech-to-text accuracy, as early transcription attempts often failed to correctly capture spoken answers
Experimenting with ElevenLabs for speech processing, which did not perform well for our use case, leading us to switch to OpenAI’s Speech-to-Text API
Accurately detecting facial expressions such as seriousness and subtle smiles, which proved difficult due to lighting, camera quality, and expression variability
Attempting to build a live, real-time conversation with the AI using LiveKit, which introduced integration complexity and stability issues
Deciding to rely on Gemini’s API for interview flow and question generation after it proved more reliable within the hackathon timeframe

Accomplishments that we're proud of

Building a complete interview simulation experience from scratch
Successfully combining voice answers, code submissions, and AI feedback
Integrating a LeetCode-style coding workflow into an interview setting
Designing a clean and intuitive UI that feels supportive rather than stressful
Implementing behavioral coaching metrics alongside technical evaluation

What we learned

Interview preparation is as much about communication and confidence as it is about technical skill
Audio handling and asynchronous flows in the browser require careful design
Security best practices (environment variables, key management) are critical even in prototypes
Thoughtful UX decisions can greatly improve how users perceive AI-driven tools