HACKTOBERFEST 2026 · WEEK 2 · VERIFIED PROOF OF CONCEPT
Real Lighthouse data.
Real open-weight AI.
This is a recorded proof of concept: GitHub Actions ran headless Chrome Lighthouse against ncodest.com, then generated actual website-advice text using a locally hosted Ollama Gemma 3 1B model.
Measured October 15, 2026, on a GitHub-hosted Linux runner. These are single-run mobile lab metrics, not current or real-user scores. The live Vercel analyzer may still need PageSpeed API quota and privately hosted AI inference configuration.
Open actual successful test and downloadable evidence artifact ↗Observed mobile Lighthouse scores
Website: https://ncodest.com/
Performance
57 /100Accessibility
78 /100SEO
100 /100Best Practices
100 /100Gemma-generated suggestions — reviewed summary
The suggestions below summarize the real Gemma output in the CI log; they are not independent proof that any specific fix is needed. The model also invented a possible service price; that price was not evidence-based and is deliberately omitted here.
Improve semantic HTML and accessible names
The captured Gemma response prioritized ARIA usage, accessible button names, and correctly labelled controls. Review the actual affected elements before applying changes.
Investigate contrast and labeling
Check text contrast, link and button labels, and relationships between visual labels and their accessible names. Confirm each Lighthouse issue manually.
Review JavaScript workload and third-party cost
Gemma flagged main-thread work and third-party scripts for investigation. Profile the page and test changes with repeat Lighthouse runs rather than assuming a root cause.
How this works
A public URL is audited with Lighthouse. Selected scores and issues become a structured prompt for an open-weight language model running in Ollama, producing an advisory report that a freelancer can review and adapt. For this verified run we used the Lighthouse CLI, avoiding Google PageSpeed API quota. The deployed web UI uses a different audit endpoint, so the two environments must not be conflated.
Potential side hustle: a transparent, affordable starter audit with optional paid human-led accessibility, SEO and performance improvements. The commercial model has not been validated with paying clients.