Skip to content
All projects
Flagship · Case Study· 2026Featured

Intavue

An AI interview-prep platform with a real-time voice interviewer that reads your resume and asks live follow-ups — plus a code-grading sandbox and a multi-agent AI coach. Architected and built solo, end to end: web app, Windows and macOS desktop apps, billing included.

Full-Stack EngineeringApplied AIVoice AIReal-time VoiceMulti-Agent AILive CodingCode Execution SandboxCross-platform DesktopWeb AppSaaSResume/ATS AI
▸ Overview
Intavue is an AI interview-prep platform I architected and built end-to-end, solo — a web app plus native desktop apps for Windows and macOS. At its core is a low-latency, real-time voice pipeline driving an AI interviewer that reads your resume and asks live follow-up questions that adapt to your answers. Around it sits an in-browser code-execution sandbox that runs and grades solutions in JavaScript, Python, Java, and C++ on approach and edge cases — not just output matching — and a multi-agent AI coach that diagnoses weak concepts, then actively teaches them with diagrams, examples, and auto-graded quizzes. The product also includes a resume ATS scorer and rebuilder, cover-letter generation, and a STAR story bank. The real challenge wasn't any single feature — it was making all three pillars feel like one cohesive product across web and desktop. It runs live on a free tier and a $15/mo Pro plan, so I own the unglamorous half too: subscription billing, credit metering, auth, and whatever breaks in production.
▸ Impact & Performance
Web · Windows · macOS
Platforms shipped
4
Languages graded live
Solo, end-to-end
Built
▸ Challenges
  • 01

    Keeping voice latency low enough that a live AI interviewer feels conversational — while it's simultaneously reading a resume and reasoning about follow-ups.

  • 02

    Running and grading untrusted user code in four languages (JavaScript, Python, Java, C++) safely, on approach and edge cases rather than just output.

  • 03

    Making three distinct AI systems — interviewer, coach, and code grader — feel like one cohesive product across web, Windows, and macOS.

  • 04

    Operating a paid product solo: billing, credit metering, and production incidents all land on the same person who wrote the code.

▸ Solutions
  • A streamed, latency-budgeted voice pipeline: each stage (speech-to-text, reasoning, speech synthesis) gets a budget, and responses stream so the interviewer starts speaking while still thinking.

  • A sandboxed code-execution service with per-language runners and rubric-based grading that evaluates the approach and edge-case handling, not just stdout.

  • A shared design system and one product core with thin platform shells, so web and the desktop apps ship from the same brain.

  • Credit metering wired into every AI feature from day one, so unit economics are measured per request instead of discovered on the invoice.

▸ Results

Live in production at intavue.app on a free tier and a $15/mo Pro plan.

Shipped across three platforms — web, Windows, and macOS — by one engineer.

Live code grading in four languages with rubric-based scoring.

The complete SaaS spine — auth, subscriptions, credit metering — built and operated solo.

▸ Domains
aivoicesaas
▸ Work with me

Want something built like this?

I'm available for new projects. If you have an idea that needs a full-stack or AI engineer, let's talk.