Humberto Costagarcía
I’m Humberto Costagarcía, a Senior AI Evaluation Specialist and Prompt Engineer with 6+ years of experience across LLM evaluation, advanced prompt engineering, QA automation, and critical systems diagnostics.
About
I’m Humberto Costagarcía, a Senior AI Evaluation Specialist and Prompt Engineer with 6+ years of experience across LLM evaluation, advanced prompt engineering, QA automation, and critical systems diagnostics. I design metric-driven, rubric-based benchmarking for model training and alignment (including RLHF, RLAIF, SFT, and adversarial Red Teaming), with a strong focus on factuality, groundedness, safety, and system-level alignment.
I also build end-to-end automated testing pipelines—using Python, Selenium WebDriver, and Chrome DevTools Protocol (CDP)—to catch regressions, validate API behavior, and improve reliability through deterministic, reproducible test artifacts. I’m used to working with multi-turn evaluation workflows, building synthetic “golden” datasets, and enforcing structured output adherence (e.g., JSON/Markdown/regex) while conducting thorough hallucination and policy-compliance audits.
- 14 years' experience
- Joined October 2026
Skills· 9
Languages
- Spanish; CastilianFluent
- EnglishAdvanced
Industry experience
- Software & Internet
- Computers & Electronics
- Professional Services
- Manufacturing
- Energy & Utilities
Work experience· 4 roles
Built and executed metric-driven LLM evaluation workflows across major foundation models (e.g., GPT-4, Claude, Gemini), including multi-turn and rubric-based scoring for factuality, groundedness, conciseness, safety, and system alignment. Designed standardized multi-tier grading rubrics for RLHF and SFT pipelines and calibrated evaluation for consistency. Conducted adversarial red-teaming audits focused on prompt leaks, unauthorized jailbreak attempts, and catastrophic hallucination vulnerabilities. Curated high-fidelity synthetic golden datasets for supervised fine-tuning, especially for specialized reasoning and code generation tasks.
Benchmarked and evaluated 1,500+ foundation model outputs (e.g., GPT-4, Claude, Gemini) across complex domains including advanced math, Python code synthesis, deductive logic, and multi-turn instruction following. Engineered event-driven browser automation pipelines using Chrome DevTools Protocol (CDP) for reactive DOM tracking, asynchronous state resolution, and low-latency UI validation. Automated RESTful API integration tests to verify HTTP response schemas, status codes, and latency SLAs; directed technical diagnostics and preventative/corrective maintenance for PLCs and critical electromechanical equipment; enforced industrial safety protocols to prevent unplanned downtime.
Benchmarked and evaluated 1,500+ foundation model outputs across complex domains (advanced math, Python code synthesis, deductive logic, and multi-turn instruction following). Engineered end-to-end automated testing stacks using Python and Selenium/CDP for deterministic UI validation, reactive DOM tracking, and low-latency checks. Implemented RESTful API integration tests validating schemas, status codes, and latency SLAs. Produced high-fidelity defect logs with deterministic reproduction scripts, network payload traceability, and stack traces. Directed technical diagnostics and preventative/corrective maintenance practices for programmable logic controllers (PLCs) and industrial control/automation environments while enforcing industrial safety protocols to prevent unplanned downtime.
Education
Certifications and awards
Hire Humberto Costagarcía today
Post a job and invite Humberto, or contact them directly. No hidden commission.