InterviewMentor AI — Elite Real-Time Multimodal Interview

Master elite interviews with a real-time, role-aware AI application that watches your code, listens to your reasoning, and delivers brutal, Hiring-Committee-level feedback.

Comment

Inspiration: Landing a role at a Tier-1 tech company or high-growth startup is no longer just about solving a LeetCode problem; it’s about communication signal, STAR methodology, and handling pressure. Traditional mock interviews are expensive or robotic. We wanted to build a simulator that feels uncomfortably real—an AI that can interrupt you if you’re hand-waving, observe your Big-O analysis in real-time, and critique your non-verbal cues.

What it does: InterviewMentor AI provides a high-fidelity "mock loop." The Live Session: Using the Gemini 2.5 Live API, users engage in a low-latency voice and video conversation. A specialized "Cyber-Indigo" editor allows the AI to "see" code as it's typed. Role-Awareness: The system dynamically adapts based on role (Frontend, Backend, PM, etc.) and seniority (Junior to Staff). The Audit: After the session, Gemini 3 Pro performs a "Hiring Committee Audit," analyzing the transcript, final code, and video frames to provide a "Hire/No Hire" verdict with quantitative scoring on technical accuracy and STAR alignment.

How we built it: Frontend: Next.js 15 with React 19 for maximum performance. AI Core: Integrated the @google/genai SDK. We utilized Gemini 2.5 Live for the interactive session and Gemini 3 Pro for the heavy-lifting reasoning and post-interview data extraction. Visuals: Custom-built a "Cyber-Indigo" code editor using a regex-based tokenization engine for real-time syntax highlighting. Analytics: Recharts for radar-competency mapping and growth tracking.

Challenges we faced: Synchronizing real-time audio with code-vision frames required precise handling of the Gemini Live socket. Ensuring the "Interviewer Personas" (like the brutal 'Charon') maintained their personality without being unhelpful required deep system-instruction tuning.

What we learned: The reasoning capabilities of Gemini 3 Pro significantly outperform previous models when it comes to identifying subtle "red flags" in a candidate's communication style and code maintainability.

Built With

  • gemini-2.5-live-api
  • gemini-3-pro
  • lucide-react
  • next.js
  • prisma
  • react-19
  • recharts
  • tailwind-css
  • typescript
Share this project:

Updates