FitOS

FitOS is a sizing intelligence platform designed to reduce uncertainty when buying clothes online. Shoppers enter their body measurements once through a reusable Fit Passport, and FitOS compares them against product size charts, material stretch, fit preferences, and patterns mined from customer reviews. The result is an explainable size recommendation with a transparent fit score, a per-measurement tight-to-loose breakdown, missing-data warnings, and clear caveats when reviews suggest that an item runs small or large. Unlike systems that ask a large language model to guess a clothing size, FitOS keeps the final decision deterministic and testable. Its recommendation engine uses explicit sizing rules, category-specific measurement weights, fabric stretch handling, size-chart completeness, and bounded review influence. Gemma is used only for language-heavy tasks: extracting structured sizing signals from reviews, normalizing messy seller size charts, and generating shopper-friendly explanations. FitOS also provides a seller dashboard that identifies products with high fit risk, incomplete size-chart fields, recurring complaint areas, and representative review evidence. This gives sellers actionable insights that can help improve product listings, clarify sizing guidance, and potentially reduce avoidable returns. For the AMD Developer Hackathon, Gemma was deployed through vLLM on AMD Developer Cloud using ROCm. The AMD-hosted pipeline processed a synthetic benchmark of 240 reviews, producing structured fit verdicts and complaint-area analytics that can be imported into the FitOS application. The system also includes Fireworks, cached responses, keyword analysis, and deterministic templates as fallbacks, allowing the product demo to remain functional even when the GPU endpoint is unavailable.