Back to projects

AI interview practice room

Callback

Built in October 2026 as a personal product, first as Interview Copilot, then rebranded as Callback with a cinematic landing page. It is live and free, and it runs on each user’s own AI key.

Callback screenshot 1
Callback screenshot 2
Callback screenshot 3
Callback screenshot 4

Problem

Under interview pressure, people blank on their own experience or reach for vague, rehearsed lines instead of the concrete stories they actually have.

Objective

Let people rehearse the real interview out loud with honest, specific feedback that never invents achievements, while keeping their key, CV and answers private.

Solution

Practice mode writes role-specific questions from the job description and CV, takes a spoken answer through the Web Speech API, and returns a score out of 5, what worked, what to fix and a stronger answer built only from the user’s own material. Numbers that don’t appear in the CV are flagged instead of kept. Live mode captures tab or system audio with getDisplayMedia, splits it with voice-activity detection, transcribes it with Whisper large-v3-turbo on Groq (or OpenAI) and streams a grounded first-person answer. Everything runs on the user’s own key across Groq, Gemini, OpenAI, Claude and OpenRouter, called straight from the browser with automatic fallback to a backup key, so the server only sees anonymous usage counts. The landing page is a scroll-scrubbed film: five Seedance clips chained on storyboard frames, retimed to a constant camera speed with optical flow, upscaled 4× locally with Real-ESRGAN and exported as 170 AVIF frames.

Main features

What the product includes.

  • Questions written for the exact role
  • Spoken answers with a score, what worked, what to fix and a stronger version
  • Flags numbers that aren’t in the user’s CV
  • Live mode from call audio or microphone
  • Bring your own key across five AI providers, with automatic fallback
  • Keys, CV and answers never touch the server
  • English and French interface; Arabic questions and feedback
  • Scroll-scrubbed cinematic landing with kinetic type

Technology

Built with the right tools for the job.

Next.js 16React 19TypeScriptTailwind CSSGSAPGroqWhisperGeminiOpenAIClaudeWeb Audio APIVitest

Challenges

Balancing the project’s technical requirements with a clear, responsive experience—and keeping the implementation maintainable as the scope grew.

What I learned

The project strengthened my ability to turn a broad idea into structured features, make deliberate technology choices, and communicate complex functionality through a polished interface.

Next project

Have a product that needs thoughtful execution?

Tell me what you are trying to build, who it is for, and what success should look like.

Start a conversation