Frontend Engineer

Evaluate AI-generated frontend implementations against real-world reference websites, combining pixel-level visual review with source-code analysis. You will compare two candidate site replications, identify which better matches the reference across required states and viewport sizes, and provide evidence-based reasoning that distinguishes robust layouts from implementations that only appear correct at a single viewport.

• Download and run a reference site and two self-contained candidate implementations locally, using a 1920×1080 viewport and interacting with pages to reach all required hover, click, and scroll states.
• Compare visual fidelity state by state, including layout, spacing, box model, typography, color, borders, images, assets, stacking order, and overflow behavior.
• Review both codebases to assess construction quality, including semantic structure, responsive behavior, document flow, and whether layouts rely on fragile shortcuts such as hardcoded offsets, excessive absolute positioning, inline-style-heavy markup, undifferentiated div structures, or screenshots used in place of rebuilt sections.
• Test responsive behavior beyond the default viewport and clearly identify defects, such as navigation that breaks at 1400px despite appearing correct at 1920px.
• Write a specific justification for every preference, citing concrete visual or implementation evidence rather than providing a general impression.
• Use the task-broken option carefully, separating genuinely unworkable tasks from difficult but completable ones.

• 3 to 8 years of professional web-development experience, primarily delivering frontend or full-stack interfaces.
• Strong ability to inspect an unfamiliar page, infer how it was built, and identify implementation shortcuts or defects.
• Fluency with both modern and legacy web patterns, including semantic HTML, hand-written CSS, flexbox, grid, media queries, floats, table-based layouts, image maps, stacked navigation, and older site structures.
• Comfort predicting the structure behind a rendered interface before opening developer tools.
• Deep working familiarity with browser DevTools, including exact viewport sizing, element-tree inspection, computed-style review, and hover-state behavior.
• Command-line capability to unzip archives, run a local static server when asset paths require it, and troubleshoot broken image references.
• Enough JavaScript knowledge to read scripts and understand their effects on the DOM.
• Professional written English for detailed free-text rationales.

• A desktop or laptop capable of displaying a 1920×1080 viewport.
• Administrator rights on your own machine to install and run a local server.

• Prior RLHF, preference-labeling, model-evaluation, structured code-review, or rubric-driven comparison work.
• Pixel-perfect design-to-code experience, including agency work, design systems, or template production.
• Accessibility expertise, including ARIA, semantic landmarks, and heading hierarchy.
• Familiarity with common LLM code-generation failure patterns.
• Experience with web scraping, site archiving, or DOM parsing.
• More than one successfully completed prior engagement and availability for contiguous multi-hour work blocks.

• Remote hourly engagement.
• Each timed evaluation unit takes approximately 2 to 3 hours. This work requires uninterrupted multi-hour availability and is not suitable for scattered 15-minute work windows.
• Assignments include a real crawled reference website delivered as a complete zipped site tree, screenshots of its default view, and, where applicable, screenshots of states reached through hovering, clicking, or scrolling. Two candidate implementations are also provided as zipped self-contained site trees; their creators had access only to screenshots, not the original source.

$90 per hour.

• Applicants must have completed at least one prior engagement in full. New experts are not eligible for this project.
• This role is intended for practicing web developers. Backend-only, ML or data-science-only, mobile-native-only, and DevOps-only engineers are not a fit because the work requires strong UI judgment. Designers who do not code cannot assess the source-code criteria, and engineers limited to component libraries without experience authoring CSS may struggle with legacy reference pages.