Context-Engineered Reviews Architecture
A training-free framework for generating realistic, controllable synthetic review datasets for Aspect-Based Sentiment Analysis (ABSA).
CERA is a modular three-phase pipeline (Composition, Generation, Evaluation) that produces high-quality synthetic ABSA data using only context engineering and multi-agent verification — no GPU infrastructure, fine-tuning, or pre-existing embeddings required.
Developed as part of an MSc thesis at the University of Windsor.
| Repository | Description |
|---|---|
| cera | Core framework — CLI, web GUI, and the full generation pipeline |
| cera-LADy | Latent Aspect Detection evaluation framework for benchmarking generated datasets |
| cera-human-eval | Human evaluation interface for assessing synthetic review quality |
| cera-vLLM | Self-hosted local LLM server for GPU-accelerated generation |
@mastersthesis{thang2026cera,
title = {CERA: Context-Engineered Reviews Architecture for
Synthetic ABSA Dataset Generation},
author = {Thang, Kap},
school = {University of Windsor},
year = {2026},
type = {Master's Thesis}
}