Jozef T.

Key Skills

Software

No software listed

Top Subject Matter

Hospitality & Vacation Rentals

Customer Service & Guest Experience

Travel & Tourism

Top Data Types

Text

Top Task Types

RLHF

Freelancer Overview

Seven years running guest services across a 150+ property US vacation rental portfolio gives me a deep, structured library of conversational decision-making — the exact judgement layer current language models struggle with. Daily work spanned written, phone, and in-person channels: pre-arrival coordination, after-hours emergencies, complaint triage, refund-vs-credit-vs-decline decisions, owner escalations, and multi-property crisis coordination during storms and peak periods. That experience translates directly into high-quality conversational training data, RLHF rating on multi-turn customer interactions, and red-teaming AI customer service agents with the realistic adversarial inputs guests actually send — emotional manipulation, escalation threats, social-engineering attempts to extract upgrades or refunds.What sets me apart is the combination of frontline operator depth with current AI-supervision experience. In my present role at Pitchup I run rubric-based content review and high-volume outreach across six European markets using human-in-the-loop AI tooling, which means I spot model failure modes (hallucinated detail, over-apologising, missed escalation triggers, policy drift) as a matter of daily practice. I bring underrepresented domain coverage — vacation rental operations, PMS workflows, OTA dynamics, HNW client communication standards, US Southeast coastal market specifics — and I'm comfortable working from detailed rubrics, edge-case taxonomies, and structured evaluation frameworks.

Labeling Experience

Visual Content Coordinator — rubric-based daily review and quality checks of listings using human-in-the-loop AI tooling

HumanaticText

Reviewed campsite listing content daily using rubric-based criteria to ensure quality and correct classification of listing attributes. Applied structured, human-in-the-loop review practices to catch issues like hallucinated details, broken localization, off-tone phrasing, and template drift before publication. Performed iterative validation of AI-generated or drafted content against annotation/review standards used for downstream moderation-style workflows. • Rubric-based review of listings for structured quality control • Human-in-the-loop supervision and quality checking of AI-written content • Identification and reporting of AI failure modes (hallucination, localization, tone, drift) • High-volume, multi-market outreach content vetting

2025 - Present

Operator — iTrip Vacations: applied hospitality judgement to AI training/rating concepts (RLHF and response evaluation)

Don't discloseTextRLHF

Created and evaluated conversational hospitality scenario responses by applying real-world guest-services decision logic to model training and assessment concepts. Produced paired examples conceptually aligned to user messages and appropriate assistant actions including escalation decisions, liability awareness, and de-escalation under ambiguity. Designed rating dimensions for operationally relevant correctness, policy alignment, tone calibration, and identification of hospitality-specific failure modes. • Scenario-based response evaluation for factual accuracy and policy alignment • RLHF-style rating across tone, escalation judgment, and liability awareness • Failure-mode detection (over-apologizing, over-promising, hallucinated details, urgency misreads) • Adversarial/red-teaming prompt concepts for guest-service boundary conditions

2018 - 2024

Education

Work History