YEETCODE

Inspiration

We love solving algorithmic problems, but we realized that real interviews are not just about solving a question, they are about solving it under pressure, time constraints, and continuous evaluation through cross-questioning and live coding. While practicing ourselves, we often struggled to find partners for mock interviews or coordinate schedules, which made consistent preparation difficult.

This led us to ask: what truly makes an interview challenging, and can we automate that experience? Inspired by this gap, we built Yeetcode, an AI-powered coding interview platform designed to simulate realistic technical interviews by combining time pressure, dynamic questioning, live coding evaluation, and structured feedback to help candidates prepare more effectively.

What it does

Yeetcode is an AI-powered coding interview platform that simulates realistic technical interviews. Candidates select a difficulty level (easy, medium, or hard), each with defined time constraints, and an AI interviewer introduces the problem and conducts the session through dynamic questioning and live evaluation.

The platform analyzes the interview conversation to assess problem understanding, solution approaches (brute force and optimal), complexity analysis, and execution. It also tracks behavioral warnings to mimic real interview conditions. At the end, Yeetcode generates a structured report with rubric-based scoring, strengths and weaknesses, and actionable feedback to help candidates improve their performance under real-world constraints.

How we built it

We built Yeetcode using Next.js for the frontend and Node.js for the backend to manage interview sessions and real-time evaluation workflows. Gemini powers the AI interviewer, handling problem explanation, dynamic questioning, and generating structured evaluation reports. ElevenLabs provides real-time voice synthesis to create a natural interview experience, while TwelveLabs enables video-based proctoring by detecting signals such as mobile device usage or the presence of additional people.

Together, these components create an end-to-end automated coding interview simulation.

Challenges we ran into

One of the main challenges was synchronizing the interaction flow between the student and the AI interviewer to create a natural, realistic interview experience. Ensuring smooth turn-taking, maintaining conversational context, and balancing real-time evaluation without interrupting the user required careful orchestration between frontend, backend, and AI responses.

Another major challenge was implementing near real-time proctoring. We needed to efficiently capture video frames, process them in chunks, and send them to TwelveLabs for analysis while minimizing latency and avoiding performance bottlenecks. Achieving reliable detection without disrupting the interview flow required optimization of streaming and asynchronous processing.

Accomplishments that we're proud of

We’re proud of building a fully automated AI-driven coding interview experience that goes beyond simple question generation by combining conversational AI, real-time evaluation, voice interaction, and video-based proctoring into a single platform. Successfully orchestrating multiple AI systems — Gemini for interviewing and evaluation, ElevenLabs for realistic voice interaction, and TwelveLabs for behavioral monitoring into a smooth, real-time workflow was a major achievement.

We also developed a structured rubric-based grading system that provides meaningful feedback, helping candidates understand not just whether they solved the problem, but how effectively they performed under real interview constraints.

What we learned

Through building Yeetcode, we learned how to design and orchestrate multiple AI systems into a cohesive real-time application. We gained experience managing conversational flow between users and AI, balancing latency with responsiveness, and integrating multimodal components like voice interaction and video-based proctoring. This project deepened our understanding of building evaluation systems using structured rubrics and highlighted the challenges of creating fair, consistent, and realistic AI-driven assessments.