Case StudyFounder, Full-Stack & AI Engineer

Site Mechanic.

A production web-diagnostics SaaS that scans any website across SEO, performance, accessibility, security, and typography — with Claude-powered strategic recommendations.

Client
Independent Product
Year
2026
Tools
Node.js, Express, Claude API, Puppeteer, Lighthouse, Socket.IO, PostgreSQL, Stripe, Kubernetes, Docker

Overview

Site Mechanic is a full website-diagnostics platform: enter any URL and it runs a coordinated scan across SEO, performance, accessibility, security, and typography, then distills the results into a single weighted score with prioritized fixes. It is the most production-hardened project in my AI portfolio — roughly nine months of continuous development, from first commit to a live, monetized product.

How It Works

The scanning engine coordinates ten specialized analyzers — font discovery, Lighthouse desktop and mobile audits with Core Web Vitals, best practices, WCAG accessibility, font licensing, cross-browser checks, real-user monitoring, and competitive benchmarking — built on Puppeteer, Playwright, and Lighthouse. Scan progress streams to the browser in real time over Socket.IO, and finished reports export to PDF.

The AI Layer

A Pro-gated recommendations feature sends scan results to Anthropic's Claude, which returns strategic, plain-language advice tailored to the site's weakest areas. Claude also powers font-pairing analysis. The AI layer is feature-flagged and degrades gracefully without an API key, and an admin panel tracks per-model AI spend with a configurable model picker (Haiku for cost, Sonnet as default, Opus for depth).

Engineering Highlights

  • Server-side entitlements system gating features by plan, backed by Stripe subscriptions, day passes, and one-off report purchases
  • SQLite-to-PostgreSQL migration executed mid-flight with parallel test suites for both engines
  • Kubernetes deployment manifests with autoscaling and network policies, Prometheus metrics, and GitHub Actions CI/CD
  • Custom i18n layer across eight locales with no frontend framework — fast, dependency-light vanilla JS

Outcome

A live, revenue-ready SaaS with 190+ commits of sustained iteration — proof of carrying an AI-augmented product from idea through architecture, hardening, and operations.

Key Outcomes

  • 10-analyzer weighted scanning engine producing a 0–100 score and letter grade per site
  • Claude-powered AI recommendations and font-pairing analysis, gated by a server-side entitlements system
  • Real-time scan progress streamed over WebSockets with PDF report export
  • Full production ops: Kubernetes manifests, Prometheus metrics, CI/CD, and an 8-locale i18n layer
  • Stripe monetization with subscriptions, day passes, and single-report purchases