Skip to content
 
 

Repository files navigation

AdaptIQ — Adaptive Learning Intelligence

🏆 1st Place — Claude NJIT Hackathon 2026

🚀 Live Demo

AdaptIQ is a real-time, multimodal AI assistant that analyzes your body language, facial expressions, and voice during mock interview sessions to provide adaptive, empathetic support. It detects when you're confused, disengaged, or overwhelmed — and responds with intelligent interventions tailored to your cognitive profile.

Built for the Economic Empowerment and Education track, AdaptIQ is designed to level the playing field for neurodivergent job seekers. 1 in 5 Americans are neurodivergent, yet standard virtual interview formats rarely account for how differently people process and communicate information.


Demo

No install needed — runs entirely in the browser.
Visit adapt-iq-dun.vercel.app, allow camera/mic access, select a cognitive profile, and start a mock session.


What It Does

AdaptIQ runs a continuous Perceive → Process → Act loop:

  1. Perceives — captures webcam and microphone input, extracting facial landmarks, eye gaze, head pose, and audio features locally on your device
  2. Processes — sends semantic metadata (not raw video) to Claude via API for reasoning about your current state
  3. Acts — delivers real-time interventions, adjusts support style, and surfaces live feedback on a dashboard

Features

  • 🎭 Multimodal perception — face mesh tracking, eye gaze estimation, head pose analysis, blink rate, and audio processing
  • 🧠 Claude-powered reasoning — interprets your biometric state and decides on empathetic, context-aware interventions
  • 👤 Three cognitive profiles — tailored modes for ADHD, Anxiety, and ASD with different intervention styles and sensitivity thresholds
  • 📊 Live biometric dashboard — real-time metrics including Gaze Deviation Score, Head Pose Drift, Vocal Energy Spread, Speech Rate, and more
  • 🎯 Session scoring — end-of-session breakdown across Eye Contact, Head Stability, Vocal Confidence, and Speech Clarity with an overall grade
  • 🔔 Smart interventions — non-intrusive overlay cards triggered when anomalies are detected (e.g. frantic eye movement, long silences)
  • 🔒 Privacy-first — only semantic metadata is sent to the cloud; raw video never leaves your device

Cognitive Profiles

Profile Focus Intervention Style
ADHD Attention & engagement High-energy, frequent micro-breaks, gamification cues
Anxiety Calm & regulation Breathing prompts, reduced visual noise, reassuring check-ins
ASD Structure & predictability Consistent pacing, explicit transitions, clear progress markers

Metrics Tracked

Biometric Signals

  • Gaze Deviation Score (GDS)
  • Head Pose Drift (HPD)
  • Blink Rate Analysis (BRA)
  • Vocal Energy Spread (VES)
  • Speech Rate (SR)

Derived Scores

  • Off-Screen Ratio (OSR)
  • Engagement Time (ET)
  • Cognitive Engine Score (CES)
  • Silence/Latency Ratio (SILR)
  • Pitch Variance Score (PVS)

Tech Stack

Layer Technology
Face & Gaze Tracking MediaPipe Face Mesh, face-api.js
Audio Analysis Meyda.js (real-time audio features)
AI Reasoning Claude Sonnet (Anthropic API)
Frontend Vanilla HTML/CSS/JS, Chart.js
Hosting Vercel

Project Structure

├── frontend/         # UI components and dashboard
├── perception/       # Face mesh, gaze, and audio processing
├── integration/      # API orchestration and Claude integration
├── demo/             # Demo assets
├── PRD-details/      # Product requirements and design docs
├── index.html        # Main AdaptIQ dashboard
└── projectplan.md    # Architecture and implementation roadmap

Getting Started

Prerequisites

  • A modern browser (Chrome recommended)
  • Webcam and microphone
  • Anthropic API key

Run Locally

git clone https://github.com/YOUR_USERNAME/adaptiq.git
cd adaptiq

# Serve with any static server
npx serve .

Then open http://localhost:3000.


How It Works

All sensor data is abstracted into a lightweight JSON state before being sent to Claude:

{
  "gazeDeviation": 2.3,
  "headPoseDrift": 12.5,
  "blinkRate": 4,
  "speechRate": 85,
  "offScreenRatio": 0.3,
  "profile": "adhd"
}

Claude interprets this state and returns either a natural language intervention or triggers a UI action — like pausing the session or surfacing a breathing prompt.

Target latency: < 500ms end-to-end.


Built By

Built at the Claude NJIT Hackathon 2026 — themed around social impact inspired by Anthropic CEO Dario Amodei's essay Machines of Loving Grace.


License

MIT

About

Claude NJIT Hackathon 2026

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages