CipherAI

AI-powered interview practice tool that helps job seekers master behavioral questions through video-based practice. It delivers interview feedback on what you say & how you say it, so you can improve.

Comment

CipherAI — Your AI-Powered Behavioral Interview Coach

The Problem

The market for behavioral interview analysis is dominated by enterprise platforms—such as HireVue, Talview, Modern Hire, MyInterview, and VidCruiter. These tools are built strictly for recruiters, not applicants. They automate screening and scoring to optimize hiring workflows, leaving candidates on the receiving end of an opaque black box with no fair way to prepare.

The few applicant-facing "practice interview" products that do exist focus narrowly on basic communication coaching or simple transcript reviews. None of them provide real-time, data-driven scoring of both what you say and how you say it—the complete picture that actually determines whether you pass.

CipherAI flips the equation.

The Solution

CipherAI is an all-encompassing behavioral interview practice platform that empowers job seekers. It allows users to practice common behavioral interview topics, receive AI-powered verbal and non-verbal analysis, and act on specific, actionable feedback—all in one place.

🛠️ How We Built It

Onboarding & Resume Intelligence

Users begin by optionally uploading their résumé as a PDF. The backend parses it and uses Google Gemini to extract the most relevant practice topics for that specific candidate, pre-populating their setup flow. Returning users have their résumé suggestions cached so onboarding is instant.

Topic & Session Setup

Users select up to 3 of the most common behavioral interview topics (e.g., Leadership, Conflict Resolution, Teamwork). From there, they can fully customize their session:

  • Number of questions: 1–5
  • Difficulty: Easy / Medium / Hard
  • Response time limit per question: None, 1–3 min
  • Prep time per question: None, 15 s, 30 s, 1 min

Additionally, users have the option to choose tailored topics extracted directly from their resume, which are slightly more challenging and career-specific.

Question Generation

Using the selected topics and difficulty, Google Gemini (gemini-3.1-flash-lite) dynamically generates concise, realistic behavioral interview questions that authentically mirror real hiring processes.

Recording & Transcription

Responses are recorded directly in the browser using the MediaRecorder API and uploaded to Google Cloud Storage. The audio is extracted via ffmpeg and transcribed using ElevenLabs Scribe, which is tuned to capture natural speech patterns accurately.

Verbal Analysis

The transcript is processed by Google Gemini with a carefully structured prompt that produces scored, sectioned feedback across three distinct dimensions:

Dimension Score What's Evaluated
Relevance & Content /5 Did the response directly address the question? Was it substantive?
Clarity & Communication /5 Was the delivery clear, well-paced, and easy to follow?
Structure /5 Was a STAR framework used? Was the response logically organized?

These scores combine into an overall Verbal Score out of 5, accompanied by specific strengths and actionable improvement points.

Non-Verbal Analysis

Using three Google MediaPipe models running directly in the browser—Face Landmarker, Hand Landmarker, and Pose Landmarker—the platform tracks raw landmark coordinates in real time. It applies custom scoring algorithms across four dimensions:

Dimension Score What's Tracked
📏 Spatial Distribution /5 Eye-to-eye pixel distance to determine optimal face-to-camera positioning.
👋 Hand Gestures /5 Landmark movement delta over 30-frame rolling windows to score gesture frequency.
👁️ Eye Contact /5 Iris position relative to eye corners to detect camera-directed gaze.
🧍 Posture /5 Shoulder alignment and height differential to detect slouching and lateral lean.

These metrics generate a comprehensive Non-Verbal Score out of 5.

Scoring

Every session produces a complete, per-question report rendered as structured markdown. This appears as expandable cards on the feedback page, showcasing the full AI analysis alongside the transcript.

Relevant Job Matching

After reviewing their feedback, users can select "Find Relevant Jobs". The platform queries the JSearch API for live job listings matching the practiced topics and passes the results to Gemini. The AI rates each posting for candidate relevance, surfacing the most aligned opportunities first.

🧱 Tech Stack

Layer Technology
Frontend React 19 + TypeScript + Vite
Backend Node.js + Express + TypeScript
Auth Firebase (Google Sign-In)
Video Storage Google Cloud Storage
Speech-to-Text ElevenLabs Scribe v1
AI Analysis Google Gemini (gemini-3.1-flash-lite)
Non-Verbal Tracking Google MediaPipe (Face, Hand, Pose Landmarkers)
Job Search JSearch API via RapidAPI
Database MongoDB
Deployment Vercel (frontend) · Render (backend)

🎯 What Makes CipherAI Different

  • Applicant-first: Built entirely for the person preparing, not the recruiter screening.
  • Holistic scoring: Evaluates both verbal content and real-time non-verbal analysis in a single session.
  • Personalized from the start: Smart résumé parsing automatically surfaces the most relevant topics for each user.
  • Fully customizable: Complete control over difficulty, question count, response window, and prep time.
  • Actionable output: Every session concludes with targeted scores, constructive feedback, and direct jobs to apply to.

Built With

Share this project:

Updates