Careful human judgment for safer, clearer AI.
I'm Tutsirai Samantha Puwai. I rate and write AI responses in clear English, with real-world grounding in patient care and human services. That means I notice when an answer about health, caregiving or hardship is unsafe, unkind or simply unclear.
How I help train better AI
Remote, flexible work evaluating and creating data for AI models, built on my education, languages and professional experience.
Response rating (RLHF)
Comparing and ranking AI answers against detailed guidelines, with clear written reasons for every judgment.
Learn moreEnglish writing quality review
Checking AI answers for grammar, clarity, tone and consistency, and rewriting them to read naturally.
Learn moreHealth & caregiving review
Checking answers about patient care, hygiene, mobility and comfort, and flagging unsafe or inaccurate advice.
Learn moreSafety & sensitive topics
Evaluating how AI handles crisis, wellbeing, food insecurity and social-services questions, guided by human-services ethics.
Learn moreData labeling & QA
Accurate tagging, classification and record checking, built on experience with data entry, audits and privacy rules.
Learn morePrompt & answer writing
Writing realistic user prompts and model-quality reference answers in clear, natural English.
Learn moreExperience that maps directly onto AI data work
AI training rewards the same habits that patient care and human services demand: follow the protocol exactly, record precisely, protect privacy, and notice when something is unsafe.
See how each role transfersGrounded health judgment
Hands-on experience with vital signs, wound care, infection control and patient mobility helps me spot health advice that is inaccurate or unsafe.
Ethics for sensitive topics
Human-services training (person-in-environment, the NASW Code of Ethics) guides how I judge answers on crisis, hardship and wellbeing.
Precision and confidentiality
Years of data entry, audits and privacy-compliant record keeping, the same discipline labeling and QA work needs.
Real AI-training work, plus hands-on samples
An expert benchmark task I authored on real U.S. utility data, and interactive exercises where you can make the call yourself.
Expert task: data-centre site selection on real EIA utility data
An agentic benchmark task that asks AI agents to choose a data-centre site from raw federal electricity data and changing instructions. It includes 41 weighted rubric criteria, 21 automated checks, and quiet traps that separate careful reasoning from plausible-looking mistakes.
Read the case studySide-by-side rating
Two AI answers on preventing bed sores. Choose the better one, then compare with my scored rationale.
Try it InteractiveSafety flagging
An AI answer to a fever-and-stiff-neck question, with each risky claim flagged by severity.
Try it InteractiveWriting quality edit
A stiff, error-prone answer about food pantry help, rewritten for clarity, warmth and accuracy.
Try itHow I approach every task
Read the guidelines
Study the rubric and examples before the first task, and keep them open.
Calibrate
Check early ratings against reference answers and reviewer feedback.
Evaluate carefully
Judge accuracy, safety, helpfulness and tone, one dimension at a time.
Explain clearly
Write short, specific reasons a reviewer can verify in seconds.
Self-check
Re-read before submitting, and flag unclear guidelines instead of guessing.
Have an AI training or evaluation project?
I'm open to remote rating, writing and review work in English, with flexible hours on U.S. Eastern time.