CALLSCORE

Transkripsi. Skor. Tingkatkan.

AI-powered call center quality assurance platform that automatically transcribes agent-customer calls and scores performance across 7 KPIs using LLM evaluation.

Features

• Auto Transcription — Multi-speaker diarization with timestamps, powered by Gemini Flash
• AI Quality Scoring — Automated evaluation across 7 performance criteria
• Score Breakdown — Per-criteria scoring (Opening, Product, Closing, Age Check, Data Input, Legal, FAQ)
• Call History — Track and compare scores across multiple recordings
• Export — Download transcripts (.txt, .srt) and scores (.csv)
• Audio Player — Click timestamps to jump to specific moments
• Indonesian Language — Full UI and scoring notes in Bahasa Indonesia

Scoring Criteria

Criteria	Max Score	What it measures
Opening	2	Agent introduces themselves, mentions customer name, asks permission
Product	16	Clear and accurate product/benefit explanation
Closing	16	Correct closing phrase usage
Age Check	16	Customer age verification
Data Input	16	Correct data collection on positive response
Legal	16	Legal agreement read and acknowledged
FAQ	16	Clear answers to customer questions

Tech Stack

• Frontend: SvelteKit, TailwindCSS
• AI/LLM: Google Gemini 2.5 Flash (with 2.0 Flash fallback)
• Audio Processing: FFmpeg
• Deployment: Docker, Fly.io
• Features: Model retry with exponential backoff, structured JSON output, history persistence

Getting Started

git clone https://github.com/Fahlevi20/transcribe-analysis-gemini.git
cd transcribe-analysis-gemini
npm install

Create .env:

GOOGLE_API_KEY=your_api_key_here

Run:

npm run dev

Docker

docker-compose up

Architecture

Audio Upload → FFmpeg preprocessing
    → Gemini Flash (transcription + speaker diarization)
    → Structured Transcript (timestamps + speakers)
    → Gemini Flash (scoring evaluation - 7 KPIs)
    → JSON Score + Notes + Donut Chart visualization

License

MIT