QAI is an AI-powered evaluation platform that automates the analysis of customer experience quality assurance conversations using personalized, dynamic metrics. By streaming or uploading calls, QAI identifies "moments that matter" in real-time, eliminating manual scrubbing and subjective scoring.
Acting as an intelligent co-pilot, QAI listens alongside your team to provide:
- Live Agent Support: Instant, optimal solutions for handling distressed customers.
- Automated Grading: Real-time call evaluations based on your specific QA rubric.
- Manager Command Center: A comprehensive suite of tools for monitoring, logging, and coaching agents at scale.
- QAI can 10x the amount of calls a manager can analyze.
- Currently, Only 1-4% of calls are analyzed.
- 80% of customers consider the experience a company provides as important as its product and services.
- 80% of customer service organization are expected to make the jump for AI tools by 2026.
- 54% of consumers believe that customer experience at most companys need major improvements.
QAI turns quality assurance from a manual, reactive process into a fast, consistent, and coach-driven workflow—helping teams improve agent performance and customer experience at scale.
| Feature | Description |
|---|---|
| Conversation-Aware QA | Analyze full customer service calls and automatically detect good, bad, uncertain, and needs improvement moments across the conversation |
| Tone & Sentiment Detection | Track emotional changes such as frustration, hesitation, and relief in real time or post-call using voice and language signals |
| Interactive Audio Timeline | Visual timeline with markers that let quality coaches jump directly to key moments instead of listening end-to-end |
| Speaker-Labeled Transcripts | Separate agent and customer dialogue with timestamps for faster review and context |
| Rubric-Based Scoring | Score agent performance using the company’s custom QA rubric and marking scheme |
| Explainable QA Reports | Generate detailed QA reports with scores, transcript evidence, and timestamps to support consistent evaluations |
| Real-Time Agent Coaching | Surface live suggestions during calls to improve tone, empathy, and objection handling |
| Sales Guidance & Techniques | Provide contextual sales tips and conversation strategies based on what the customer is saying |
| Synthetic Call Generator | Create realistic AI-generated customer service calls with emotional progression for testing and demos |
| Scalable QA Workflow | Review more calls in less time while maintaining consistency across reviewers |
- Next.js — Application framework for fast, scalable web experiences
- React — Interactive UI for timelines, transcripts, and QA dashboards
- TypeScript — Type-safe data models for QA outputs and API contracts
- Tailwind CSS — Modern, responsive styling with consistent design
- OpenAI API — Contextual analysis for tone detection, QA classification, scoring, and coaching insights
- Soniox API — Real-time and batch speech-to-text with speaker separation and timestamps
- Node.js / Express — Core backend for audio ingestion, QA orchestration, and API endpoints
Make sure you have the following installed and set up:
- Node.js 18+
- npm or pnpm
- OpenAI API key (used for analysis, tone detection, coaching, and live feedback)
- Soniox account (used for real-time and batch transcription) Get a key at: https://console.soniox.com
# Navigate to the project directory
cd hackhive26
# Install dependencies
npm installCreate a .env.local file in the project root and add the following variables:
# AI / Transcription
OPENAI_API_KEY=your_openai_key
SONIOX_API_KEY=your_soniox_keynpm run dev
hackhive26/
├── app/ # Next.js App Router
│ ├── api/ # API routes
│ │ ├── analyze/ # AI analysis endpoints
│ │ ├── transcribe/ # Audio transcription
│ │ └── soniox/ # Soniox (tone, coaching, live feedback)
│ ├── analytics/ # Analytics page
│ ├── dashboard/ # Dashboard page
│ ├── live/ # Live call page
│ ├── live-call/ # Live call page (alternate route)
│ ├── qa/ # QA review page
│ ├── page.tsx # Home page
│ ├── layout.tsx # Root layout
│ └── globals.css # Global styles
│
├── components/ # Reusable React components
│ ├── ui/ # shadcn/ui components
│ ├── analysis-panel.tsx # Analysis panel
│ ├── app-sidebar.tsx # App sidebar
│ ├── upload-zone.tsx # Upload zone
│ ├── waveform-timeline.tsx # Waveform timeline
│ └── theme-provider.tsx # Theme provider
│
├── hooks/ # Custom React hooks
│ ├── use-mobile.ts
│ └── use-toast.ts
│
├── lib/ # Utilities & helpers
│ └── utils.ts # Helper functions
│
├── public/ # Static assets
│
└── styles/ # Stylesheets
└── globals.css
- Inconsistent QA interpretations during subtle tone shifts
- Overconfident classifications in ambiguous or low-confidence moments
- Emotional tone changes gradually across conversations, making single-utterance analysis unreliable
- Latency when running real-time transcription and QA analysis simultaneously
- Heavy scoring logic not suitable for live call processing
- Merge conflicts caused by rapid prompt and UI iteration
- Inconsistent behavior across branches during experimentation
- Early QA outputs felt like a black box for quality coaches
- Difficulty validating AI decisions without emphasizing explainability
- OpenAI — Tone detection, contextual analysis, QA scoring, and live coaching insights
- Soniox — Real-time and batch speech-to-text transcription with speaker separation
Team:
- Abinan Suthakaran
- Adam Marcelo
- Jordan Earle
- Hamzah Al-Hamadani