A toolkit for generating high-quality synthetic contact center transcripts (JSON) to demonstrate Google Cloud CCAI Insights features like Topic Modeling, Smart Highlights, and Quality AI.
This project simulates VeloFit, a fictional fitness wearable company, generating realistic customer service calls including "Happy Path" returns, policy disputes, and complex multi-agent escalations.
1. Setup Environment
export GOOGLE_API_KEY="your-key"2. Run Generator
# Generates 10 synthetic calls
uv run python -m src.synth_data.main --count 103. Check Output
Files will be generated in data/synthetic_transcripts/.
src/synth_data/: Core Package. Contains the modularized generation logic.main.py: Entry point.config.py: Profiles, personas, and product data.generators.py: Prompt engineering and LLM iteration.
src/ops/: Operational utilities (splitting files, data enrichment).src/validation/: Schemes validation scripts.src/reporting/: Dashboard and stats generation.src/legacy/: Deprecated scripts (e.g., Generic profile).
- PROCESS.md: Detailed Order of Execution and workflow guide.
- SETUP_GUIDE.md: Manual configuration steps for Google Cloud Console.
- VELOFIT_SIMULATION.md: Details on the simulation logic (Time-phased error rates, personas).
- Valid CCAI JSON: Strictly compliant with Google Cloud ingestion formats.
- Time-Phased Logic: Simulates "Steady State", "Launch Chaos", and "Learning Curve" phases based on call dates.
- Dynamic Personas: Agents and Customers behave according to specific personas (e.g., "Aggressive Customer", "Rookie Agent").
- Escalation Simulation: Automates realistic transfers from Tier 1 agents to Supervisors.