LiveIndustry-Specific

Ensemble

Thesis tool: stylometry + model ensemble for AI-assisted writing detection.

Ensemble interface

The Problem

Single detectors miss a lot. Ensemble combines stylometry with a few transformer checks and a small meta-model so the thesis could compare approaches on real drafts.

Architecture

API Gateway + Lambda for stylometric analysis

SageMaker Async Inference for GPU models (Binoculars, Fast-DetectGPT)

Multi-model calibration via stacking meta-learner

Engineering Decisions

Why stack models?

Each method fails on different inputs. Combining them beat any one model in testing, especially on paraphrased text.

Why SageMaker async?

Lambda can't fit 7B models or a GPU. Async endpoints spin down when nothing's queued.

Highlights

Full document uploadGPU jobs scale to zero when idleBuilt for research, not as a consumer product

Stack

Next.jsSageMakerStylometric AnalysisHuggingFace