Website Quality Scorer
How One ML Pipeline Scores Any Site in Seconds (XGBoost + SHAP)

Client
Challenge
Every business owner, marketer, and agency founder has been told their website needs to be better. But "better" is vague. What specifically is broken? Is it the copy, the performance, the trust signals, or the layout? And what should you fix first? The default approach is to run a single-metric tool (PageSpeed for performance, SEO checker for keywords) and guess at the rest. But no single tool covers the full picture, and none of them tell you why a score is what it is, or what to change. Four specific problems: Scattered diagnostics. Performance, content, trust, UX, each requires a separate tool, separate report, separate login. No unified view exists. No explainability. A score of 42 tells you your site is bad. It doesn't tell you that 8.3 points of that are due to a missing hero CTA, 5.1 are from no testimonials, and another 6.2 are from slow LCP. You get a number, not a diagnosis. Bulk analysis is impractical. Running 200 prospect sites through 4 separate tools each is hours of manual work. There's no way to batch-score sites across all dimensions at once. No fix priority. Even with a full audit, you get a list of problems, not an ordered list of what moves the needle most. Fixing the wrong thing first wastes time. Aleem built a single endpoint that solves all four, crawl any URL, extract 40 features across 4 dimensions, run XGBoost + SHAP, and return a scored, explained, prioritized report in seconds.
Goal
The Website Quality Scorer is a FastAPI backend with a Next.js frontend that takes any public URL and returns a 0-100 quality score broken into UX, content, technical, and trust sub-scores, each explained by SHAP feature contributions. By the end of a single POST request: A 0-100 overall score with a tier label (Poor / Average / Good / Excellent) 4 sub-scores (0-25 each) mapped from SHAP contributions per feature group The top 12 SHAP-ranked features showing exactly what helped or hurt 5 prioritized recommendations with impact estimates ("+3 to +6 points") Elapsed time in milliseconds, model version, and the scored URL One URL in. Full diagnostic out. No dashboard login, no multi-tool workflow.
Result
The Website Quality Scorer is an ML-powered diagnostic pipeline that turns any public URL into a structured, explained assessment. Aleem built it as his Machine Learning course project, but the architecture solves a problem nearly every business faces: understanding what's wrong with their website and knowing what to fix first. The architecture, crawl, extract structured features, run a tree-based model, explain with SHAP, and generate template-driven recommendations, is not specific to website scoring. The same pipeline works for: SEO agencies scoring hundreds of prospect sites to prioritize outreach with data-backed hooks Ecommerce brands auditing product pages for conversion gaps at scale SaaS companies running weekly quality checks on landing pages after redesigns Web development agencies generating automated pre-build audits for every new client Freelance designers producing professional site assessments as lead magnets The core innovation is the SHAP feedback loop: instead of a black-box score, the client sees exactly what costs them points and what to fix first. That explainability is what separates a curiosity from a sales tool.