Evan H.

Key Skills

Software

Prodigy

Google Cloud Vertex AI

Top Subject Matter

B2B SaaS / collaboration

Text classification & taxonomy design

RLHF-style evaluation

Top Data Types

Text

Image

Video

Top Task Types

Function Calling

Prompt + Response Writing (SFT)

RLHF

Text Summarization

Text Generation

Data Collection

Computer Programming/Coding

Question Answering

Object Detection

Fine-tuning

Point/Key Point

Freelancer Overview

I’m a dedicated data trainer with 3+ years of experience creating and curating AI training data, including company-adjacent projects and ongoing side work. I’m used to following detailed guidelines, applying consistent labels, catching edge cases, and redoing work when quality bars aren’t met—the same habits you need for clean datasets and reliable model behavior. What helps me stand out is I also build and refine the workflows around the work: structured specs, clear rubrics, review steps, and lightweight quality checks so annotations stay consistent and auditable over time. I’m comfortable communicating with stakeholders, documenting decisions, and iterating when instructions or task design need to change.

Labeling Experience

Spec-driven QA & failure-mode labeling for internal Claude Code agent workflows

TextRLHF

In support of an internal agent pipeline (Director / Researcher / Builder roles) with quality hooks, I help keep outputs aligned to engineering specs and team standards. Work includes tagging recurring failure modes (misinterpreted requirements, incomplete changes, risky refactors), labeling rework vs. accept decisions, and tightening rubric-style criteria so reviews stay consistent across projects. This directly informs what gets escalated, what gets regenerated, and how the workflow stays repeatable for engineers shipping production React/Next.js and Vue/Nuxt codebases.

2025 - Present

Text moderation & policy tagging for user-generated content

TextClassification

Labeled UGC against a detailed policy guide (e.g., harassment, hate, sexual content, spam, self-harm). Applied multi-label categories and priority/severity fields; flagged ambiguous cases with brief rationales per workflow. Focused on edge cases, borderline content, and consistent application of updated guidelines through calibration notes and spot audits.

2024 - 2024

Education

Work History