Carlos Rocha

Experienced in RLHF and RLAIF evaluation workflows, producing preference data and SFT-quality prompt/response pairs. Conducts expert annotation across factuality, safety, and technical domains using structured rubrics. Contributes to rubric design for specialist tasks. Applies adversarial reasoning to red-teaming. Familiar with prompt injection attack vectors and their evaluation in safety testing contexts.

AI Training Consultant at Large remote contractor (under NDA) (2026-01 – Present)

• Expert LLM response evaluation, factuality review and rubric-based scoring; producing preference data and SFT-quality prompt/response pairs across governance, security and technical domains, evaluating ~20 responses per week across 5 task categories.
• Red-teaming model outputs and testing for prompt injection vulnerabilities; evaluating multi-step reasoning, explanations and answer consistency.

Managing Director / Technology Entrepreneur at Samurai Digital Transformation & predecessor ventures (1995-01 – 2025-12)

• 2015–2025: Governance advisory and board-level technology strategy for public-sector and enterprise clients.
• 2005–2015: Designed and deployed OTT platforms serving 1M+ users (SKY, TIM); multi-system modernisation and high-scale content delivery.
• 1995–2005: Led design, manufacturing and rollout of 200,000+ secure electronic voting machines for Brazil's national electoral infrastructure; directed independent technical audits.

Visiting Researcher at Imperial College Business School (2025-01 – Present)

• Research focus: AI governance frameworks, digital trust and accountability in high-stakes automated systems.

B.Sc. in Electronic Engineering – ITA (Instituto Tecnológico de Aeronáutica)