Guillermo G.

Key Skills

Software

Scale AI

Labelbox

Top Subject Matter

Mathematics - Analysis (Orthogonal Polynomials and Special Functions)

Physics

Scientific Research

Top Data Types

Image

Text

Document

Top Task Types

Question Answering

Text Summarization

RLHF

Transcription

Evaluation/Rating

Fine-tuning

Freelancer Overview

I am a Researcher at the Centre for Mathematics of the University of Coimbra (CMUC) and a PhD candidate in Mathematics with a background in both Mathematics and Physics. My research focuses on orthogonal polynomials, special functions, inverse spectral problems, recurrence relations, and mathematical physics. I am the author of several peer-reviewed publications, including articles in Letters in Mathematical Physics, Integral Transforms and Special Functions, Journal of Mathematical Sciences, and Physica D. Alongside my academic research, I have more than one year of experience as an AI Mathematics Expert and AI Trainer at Outlier, where I have contributed to multiple expert-level projects involving mathematical reasoning evaluation, benchmark design, proof verification, rubric development, and reinforcement-learning-from-human-feedback (RLHF) workflows. As part of the Oracle team at Outlier, I have created and evaluated mathematical content ranging from advanced high-school and undergraduate problems to graduate and research-level mathematics. My work has covered graph theory, topology, real and complex analysis, algebra, differential equations, mathematical physics, and orthogonal polynomial systems. I have designed original benchmark datasets, assessed model-generated proofs and derivations, developed scoring rubrics, identified reasoning failures and hallucinations, and contributed to the training and evaluation of frontier AI systems. A distinguishing aspect of my profile is the combination of active mathematical research, peer-reviewed publications, invited conference talks, and hands-on experience creating research-level benchmarks and evaluation datasets for large language models.

Labeling Experience

Mechanic Glen V2 – Mathematical Response Evaluation and Rubric Design

TextEvaluation/Rating

Evaluated and ranked multiple AI-generated mathematical responses according to correctness, reasoning quality, instruction following, clarity, and mathematical rigor. Tasks involved comparing competing model outputs, identifying logical errors, verifying derivations, and producing detailed justifications for rankings. Additionally, the project required designing novel evaluation rubrics and grading dimensions to measure mathematical intelligence, reasoning quality, and problem-solving ability. Particular emphasis was placed on proof verification, mathematical accuracy, and rigorous assessment of model performance.

2026 - 2026

Project El Dorado – Image-Dependent Mathematical Reasoning Benchmarks

ImageRLHF

Created advanced image-dependent mathematical reasoning benchmarks designed to evaluate multimodal AI systems. Tasks required generating original college-level problems whose solutions depended critically on information contained in diagrams, graphs, geometric constructions, network structures, and visual representations. Problems were designed to be challenging for state-of-the-art models and required multi-step reasoning. My contributions included creating original prompts, constructing ground-truth solutions, validating final answers, and verifying model failures. The majority of tasks involved geometry, graph theory, combinatorics, and visual reasoning. Strict quality-control procedures were followed to ensure uniqueness, correctness, and image dependence of every benchmark. The project specifically targeted difficult multimodal mathematical reasoning tasks beyond standard high-school curricula.

2026 - 2026

Cracked Vault – Scientific Paper Reasoning Benchmarks

TextRLHF

Created expert-level reasoning benchmarks derived from scientific research papers hosted on arXiv and related repositories. Tasks involved selecting research papers within my domain of expertise, designing challenging reasoning prompts based on the paper content, constructing verifiable ground-truth answers, and validating that multiple AI systems failed to solve the problem correctly. The project emphasized scientific understanding, research comprehension, and the creation of novel reasoning tasks inspired by current mathematical research. Many benchmarks were based on advanced topics in orthogonal polynomials, spectral theory, recurrence relations, and mathematical physics.

2025 - 2025

Phoenix – Research-Level Benchmark and Rubric Development

TextRLHF

Designed complex research-oriented benchmark tasks and evaluation rubrics intended to assess advanced reasoning capabilities of large language models. Responsibilities included creating difficult prompts, constructing detailed scoring criteria, developing positive and negative evaluation rubrics, and assessing model responses against those criteria. Many tasks involved university-level and research-level mathematics, scientific reasoning, and information retrieval. Special attention was given to creating rigorous, objective, and reproducible evaluation frameworks capable of distinguishing high-performing models from weaker systems.

2025 - 2025

Education

Work History