FixMySite

FixMySite AI turns complex website audits into clear, prioritized fixes—so teams know exactly what to improve next.

Comment

FixMySite AI was inspired by a simple problem: website audit tools can identify dozens of issues, but they often leave people wondering what to fix first and how to turn technical findings into real improvements. Performance scores, accessibility warnings, SEO checks, and security headers are useful—but they can be overwhelming for founders, small teams, and developers working quickly.

We built FixMySite AI to bridge that gap. A user submits a public website URL, and the app runs a homepage audit using Lighthouse, axe-core accessibility checks, Chromium, and passive security-header inspection. Instead of only showing raw scores, FixMySite AI organizes the evidence into a health score, prioritized issues, explanations, and actionable next steps.

For the AI experience, we use the Groq API with GPT-OSS 20B. The model receives a minimal, structured audit context—such as category scores, issue severity, evidence, and recommended actions—and returns a clear summary, a priority plan, and AI-assisted fix guidance. We made sure the AI does not replace the actual audit tools or invent scan results. Responses are validated with Zod, and the app has deterministic fallback guidance if an AI request is unavailable.

We built the product with React, TanStack Start, TypeScript, Tailwind CSS, Supabase Auth/PostgreSQL, Netlify Functions, Lighthouse, Puppeteer/Chromium, axe-core, and Groq. OpenAI Codex helped us during development with planning, implementation, refactoring, documentation, and verification; Codex is not part of the deployed runtime path.

One of the biggest challenges was making the AI useful without making it unreliable. We solved this by grounding every AI request in validated scan data, limiting the context sent to the model, requesting structured output, and keeping human review central for generated fix suggestions. Another challenge was running browser-based audits in a serverless environment, where Chromium setup, cold starts, and execution time all matter.

Through this project, we learned how important it is to separate deterministic systems from generative AI. The scanner should provide trustworthy evidence; AI should help people understand and act on that evidence. FixMySite AI is our attempt to make website quality work more approachable, practical, and actionable.

Built With

Share this project:

Updates