Juan R.

Key Skills

Software

Internal/Proprietary Tooling

Top Subject Matter

Information Technology – Systems Administration & DevOps

Artificial Intelligence – Machine Learning & Model Routing

Software Engineering – Scripting (Bash) & Architecture

Top Data Types

Text

Document

Top Task Types

Classification

Entity (NER) Classification

Text Generation

Fine-tuning

Freelancer Overview

As a systems administrator and advanced IT student specializing in Artificial Intelligence and Big Data, I possess a strong technical foundation in data preparation, structuring, and processing for algorithmic training. My hands-on experience includes fine-tuning BERT-based models for document classification tasks, which has given me a deep understanding of how precise data labeling directly impacts final model performance and reliability. Beyond data management, I excel at developing and deploying end-to-end AI architectures. I have designed and published open-source projects such as LeMoE, a lightweight Mixture of Experts (MoE) router, and I am actively working on model integration for the WatermelonD voice assistant. These initiatives require me to continuously evaluate Small Language Model (SLM) outputs, manage API routing, and optimize inference workloads using Docker and high-performance computing hardware. This blend of infrastructure administration and AI development allows me to bring a highly technical, analytical, and detail-oriented approach to AI data evaluation and training.

Labeling Experience

Binary Classifier

TextClassification

The "LeMoE" (Lightweight Mixture of Experts) project focused on developing a highly efficient routing system for Small Language Models (SLMs) and APIs. My work involved creating, labeling, and validating specialized datasets for intent classification and prompt routing. The core data labeling tasks required analyzing user prompts and accurately categorizing them by domain, computational complexity, and the optimal target "expert" model (Text Classification). To guarantee high accuracy, I implemented strict quality control measures to ensure the router could dynamically direct queries to the most appropriate model, optimizing both response quality and underlying system resources.

2026 - Present

Grape T5 Models

TextText Generation

The "Grape T5 Models" project involved the creation, curation, and comprehensive validation of a specialized dataset for fine-tuning T5 family models. The primary goal was to develop a system capable of accurately translating complex natural language instructions into executable Bash commands. Specific data labeling tasks included generating instruction-command pairs (Text Generation), covering everything from basic system utilities to advanced network and container administration tools. To ensure the highest quality, I implemented a rigorous manual technical review process, ensuring that each command was not only syntactically correct but also safe, efficient, and aligned with Linux system administration and scripting best practices.

2025 - 2026

Education

Work History