Intervue AI

Intervue is an AI assistant that helps you train and perfect your oral and written skills to get your dream internship/job! You're not quite confident about the process? No worries, let InterBlob help

Comment 2

Intervue AI – Project Story

What inspired this project

Interviews are increasingly shaped by AI, but most tools either over-automate the process or ignore what actually matters: clear thinking, structured communication, and honest feedback for candidates. Watching friends stress over technical interviews, recording themselves on their phone, and manually timing answers showed how broken the preparation experience still is.

Intervue AI started as a simple question: what if interview prep felt like a realistic, adaptive conversation instead of a static list of “Top 50 questions”? The goal was to build an AI interviewer that reacts to what a candidate actually says, asks meaningful follow‑ups, and produces feedback that is specific, measurable, and doable.

How the project works

Intervue AI is a fully agentic AI system that runs an entire mock interview autonomously. No CV or job description integration, just purely chat-driven.

User starts with one message:


"I'm preparing for a Senior Full-Stack Engineer role at level L5. Focus on React, Node, system design, and behavioral questions."

The agent takes over completely:

  1. Initial parsing

Extracts:

  • Level: L5 (Senior) → expects deep expertise, leadership, trade-offs

  • Role: Full-Stack Engineer → React frontend + Node backend + deployment

  • Skills: React, Node, system design, behavioral

  • Interview structure: 45min format (3 behavioral, 3 technical, 1 system design)

  1. Dynamic interview loop

   Agent: "Great, L5 Full-Stack Engineer. Let's start with behavioral. 

   Tell me about a time you led a major frontend refactor under tight deadlines."



   [User responds...]



   Agent: "Solid example. Follow-up: What trade-offs did you make between 

   performance and developer experience? Walk me through your React state 

   management decision."

Challenges faced

  1. Agent drift

Early agents would repeat questions or go off-topic. Fixed with:

  • Strict state machine (no free-form chat)

  • Question deduplication via embedding similarity

  • Hard timeout after 8 questions

  1. Level calibration

L5 questions kept being too easy. Solution:

  • Explicit "L5 rubric" in every prompt

  • Sample L5 answers for few-shot

What was learned

Agentic AI shines for interviews because they have natural structure: question → answer → evaluate → route. No need for complex RAG or tools—just smart state management.[1]

Level specification is critical. Saying "L5 Full-Stack" immediately sets expectations for depth, trade-offs, and leadership vs. just "implement a todo app."

Chat-only is liberating. No UI complexity, no session management headaches. Users just paste their target role/level and go.

Intervue AI proves agentic systems can deliver production-grade interview practice with <100 lines of core logic. Next: voice mode + real-time latency optimization.

Built With

  • api
  • api-endpoints
  • api-routing
  • css3
  • dom-manipulation-##-**frameworks-&-libraries**-**flask-3.0**-web-server
  • eleven-turbo-v2)
  • elevenlabs
  • file-processing-**html5**-frontend-structure-**css3**-custom-animations
  • flash
  • flask
  • flask-cors
  • gemini
  • gemini-2.0-flash-exp)
  • github
  • google
  • gradients-**javascript-(es6+)**-real-time-streaming
  • gunicorn
  • html5
  • javascript
  • pypdf2
  • python
  • render
  • requests
  • responsive-design
  • routing-**flask-cors-4.0**-cross-origin-requests-**gunicorn-22.0**-production-wsgi-server-**pypdf2-3.0**-pdf-resume-parsing-##-**cloud-platforms**-**render.com**-free-tier-hosting-&-deployment-**github**-source-control
  • scribe-v2)
  • speech-to-text
  • text-to-speech
  • webrtc-audio
Share this project:

Updates