Freddy Rojas Robles

Key Skills

Software

Remotasks

Scale AI

Other

Top Subject Matter

Red Teaming in English & Spanish

Prompt+ Response writing Culturally ( Latam+Spanish)

LLM in Spanish

Top Data Types

Document

Image

Text

Top Task Types

Prompt Response Writing SFT

Red Teaming

Text Generation

Text Summarization

Translation Localization

Freelancer Overview

I work as an AI Data Trainer and Prompt Engineer, with a strong academic background in Economics and International Business. Throughout my career, I have excelled in tasks that require both critical thinking and detailed analysis— particularly those related to data labeling and AI training. My approach revolves around enhancing the quality of AI models through the implementation of high-quality data engineering practices. One of the notable projects I led was the Safety AI Data Project where I designed questions that ensured the fairness and safety of AI regardless of geographical disparities. This experience was complemented by my involvement in the Visual Prompt Enhancement Project aimed at clarifying the responses generated by AI systems for improved effectiveness. My contribution to the field of AI is unique given my ability to leverage technical competencies towards innovative problem-solving approaches.

Labeling Experience

Safety Ai data trainer

OtherTextClassificationQuestion Answering

This was a large-scale project focused on improving the safety, quality, and accuracy of leading large-language models (LLMs) like GPT-3.5/4 and Google Gemini. My specific data labeling tasks included: 1) Red-Teaming: Creating adversarial prompts to identify and classify safety policy violations. 2) Prompt Refinement (SFT): Writing and rewriting prompts to enhance the model's instruction-following capabilities and response quality. 3) Response Evaluation: Scoring model outputs against detailed rubrics for coherence, factual accuracy, and helpfulness. Quality measures were strictly adhered to through a multi-step review process and maintaining a high inter-annotator agreement (IAA) with the rest of the team.

This was a large-scale project focused on improving the safety, quality, and accuracy of leading large-language models (LLMs) like GPT-3.5/4 and Google Gemini. My specific data labeling tasks included: 1) Red-Teaming: Creating adversarial prompts to identify and classify safety policy violations. 2) Prompt Refinement (SFT): Writing and rewriting prompts to enhance the model's instruction-following capabilities and response quality. 3) Response Evaluation: Scoring model outputs against detailed rubrics for coherence, factual accuracy, and helpfulness. Quality measures were strictly adhered to through a multi-step review process and maintaining a high inter-annotator agreement (IAA) with the rest of the team.

2023 - 2025

Safety Ai Data Projext

Scale AITextRed Teaming

Objective: Develop AI tools to monitor public discourse, ensuring compliance with cultural sensitivities and legal standards. Key Responsibilities: Curated and labeled large datasets from various communication channels, including social media and public forums. Implemented natural language processing techniques to detect and analyze sensitive topics. Worked with public agencies to deploy AI solutions that aid in proactive governance and crisis management. Outcome: Created a reliable AI monitoring system that assists public agencies in maintaining safety and appropriate public discourse, significantly reducing response times to critical incidents.

Objective: Develop AI tools to monitor public discourse, ensuring compliance with cultural sensitivities and legal standards. Key Responsibilities: Curated and labeled large datasets from various communication channels, including social media and public forums. Implemented natural language processing techniques to detect and analyze sensitive topics. Worked with public agencies to deploy AI solutions that aid in proactive governance and crisis management. Outcome: Created a reliable AI monitoring system that assists public agencies in maintaining safety and appropriate public discourse, significantly reducing response times to critical incidents.

2023 - 2024

Education

Work History