Most automated test generators only cover the "Happy Path"—the expected user behavior. They miss the Edge Cases: the statistical anomalies and boundary conditions that cause 90% of production outages.
Zeta is not just an LLM wrapper. It is a hybrid system that combines Generative AI (for creativity) with Statistical Physics (for rigor).
It calculates the Complexity Z-Score of every software requirement. If a requirement is statistically "dense" (high logic complexity) but has low test coverage, Zeta flags it as a Critical Risk.
| Feature | Standard LLM Gen | ⚛️ Zeta (Physics Engine) |
|---|---|---|
| Generation | Hallucinates generic tests | uses Gemini 2.5 with SOTA Prompting |
| Validation | None (Blind trust) | IsolationForest + Z-Score Anomaly Detection |
| Safety | Returns broken JSON often | Pydantic Type-Safety & Self-Healing |
| Output | Text descriptions | Executable Selenium/Python Code |
- Engineering Log: A deep dive into the bugs we fought (NumPy ambiguity, Data Loss) and how we fixed them. (Must Read for Engineers).
- User Guide: Step-by-step instructions on generating tests and interpreting Z-Scores.
- System Architecture: High-level diagrams of the FastAPI/Streamlit microservice topology.
- Core: Python 3.11, FastAPI, Uvicorn
- ML Engine: Scikit-Learn (Isolation Forest), NumPy (Vectorization)
- Frontend: Streamlit, Altair (Visualization)
- Infrastructure: Docker, Hugging Face Spaces
Owadokun Tosin Tobi Senior AI Engineer & Physicist