Jessy M.

Key Skills

Software

Don't disclose

Top Subject Matter

AI model output evaluation and benchmarking

LLM document extraction and accounting invoices

AI model behavioural evaluation

Top Data Types

Text

Document

Top Task Types

Red Teaming

Evaluation/Rating

Computer Programming/Coding

Prompt + Response Writing (SFT)

AI Trainer Overview

I evaluate AI output and build the harnesses that do it at scale. On a paid contract for a TikTok-affiliated client I built the task set and scored 35 benchmark cases against a defined rubric, delivered across two approved milestones. I built Golden Arena, a five-game harness with a behavioural index that ranks models on how they actually play rather than on what they claim. Before an LLM document-extraction pipeline went anywhere near real accounting books, I ran a 25-invoice reliability battery against it and held the release until it passed. I work to written rubrics and stop-conditions, and I report an inconclusive result as inconclusive. I am native in English and French, so I can evaluate and compare output in both. My background is project-led rather than academic: two years building production TypeScript, Node and PostgreSQL systems, with open-source tooling published to npm and the official Model Context Protocol registry.

Labeling Experience

Golden Arena behavioural evaluation harness

Comparative behavioural scoring of AI models through five adversarial games and a Behavioural Index.

2026 - Present

LLM document pipeline reliability evaluator

Don't discloseDocumentEvaluation/Rating

Evaluated document-extraction outputs across 25 invoice cases for reliability and correctness.

2026 - Present

AI evaluation benchmark contractor

Don't discloseText

Scored 35 benchmark cases against a defined rubric as part of an AI evaluation/crowdtesting contract.

2026 - Present

Education

Work History