Arriaga T.

Key Skills

Software

Mercor

Don't disclose

Top Subject Matter

AI safefy at Mercor

mathematical reasoning at BUAP

LLM interpretability

Top Data Types

Text

Top Task Types

Red Teaming

Fine-tuning

RLHF

Evaluation/Rating

Data Collection

Transcription

Text Summarization

Computer Programming/Coding

Prompt + Response Writing (SFT)

Text Generation

Question Answering

Freelancer Overview

AI Researcher — Mathematical Reasoning & Alignment (Freelance). Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Mercor and Don't disclose. Education includes Bachelor of Science, Benemérita Universidad Autónoma de Puebla (BUAP) and Diploma, Benemérita Universidad Autónoma de Puebla (BUAP) (2025). AI-training focus includes data types such as Text, Computer Code, and Programming and labeling workflows including Red Teaming and Fine-tuning.

Labeling Experience

AI Researcher — Mathematical Reasoning & Alignment (Freelance)

MercorTextRed Teaming

Directed theoretical auditing of complex mathematical proofs generated by LLMs to ensure logical, topological, and algebraic consistency across large datasets. Engineered edge-case prompt injection attempts using logical paradoxes and computational complexity theory to probe the formal boundaries of production AI models. Framed preference optimization for RLHF using a thermodynamic free-energy minimization perspective to improve convergence behavior during training.

2025 - Present

Computational Physics & ML Researcher

Don't discloseFine-tuning

Built machine learning pipelines to analyze large-scale physics simulation records using Monte Carlo methods for particle cascade modeling. Benchmarked convolutional and transformer architectures against SO(3)-equivariant networks for particle classification, using physical symmetry embeddings to reduce sample complexity and improve sensitivity. Applied manifold learning techniques (t-SNE, UMAP) to extract invariant macroscopic observables from high-dimensional detector-state representations.

2025 - 2025

Education

Work History