SunGrid Living
A 47-page hurricane-prep resource with its own dataset, calculators and AI-search layer
Visit the live site →Role: Solo — design, build, content architecture, SEO, infrastructure
The problem
Florida loses power every hurricane season, and the advice online is either vague ("get a generator") or wrong. The specific question people actually ask — will my fridge survive this, and what do I need to buy — had no trustworthy answer written down. I wanted to find out whether a small, rigorously sourced site could become the thing AI assistants quote when someone asks that question.
How I built it
Built the data layer first
One JSON file is the single source of truth for every product, spec and price band. The comparison tables, the picks pages, the public dataset downloads and the cost-per-watt-hour column all render from it — so a number can never disagree with itself across pages.
Made the tools refuse to lie
The runtime calculator declines to answer when an answer would mislead. Ask it to run a 500W air conditioner on a 300W battery and it says the unit cannot do it at all, instead of returning a runtime that would send someone shopping for the wrong thing.
Published the methodology
Every formula, derating assumption and data source sits on a public page, including a plain statement of what has not been tested. Accuracy is the product.
Wired the machine-readable layer
A full Schema.org graph, an llms.txt index, an openly licensed dataset with CSV and JSON downloads, and a citation block — everything an AI retrieval system needs to quote the site correctly.
Shipped email capture with no third party
A Cloudflare Pages Function writing to Workers KV, with a token-protected export endpoint. No Mailchimp, no monthly bill, no customer data leaving the account.
The outcome
47 pages live on Cloudflare's edge at zero monthly hosting cost. An adversarial audit — parallel agents whose only instruction was to disprove every numeric claim on the site — surfaced 86 issues before any of them reached a reader, and all were fixed. A scheduled monthly job now re-verifies all 34 product listings, because product links rot within weeks and stale data is the fastest way to lose a reader's trust.
The interesting engineering was not the pages. It was making a content site behave like a maintained system: one source of truth, automated verification, and tools that decline to answer rather than answer wrongly.
Technical stack and details
A Florida backup-power research site built to be the source AI assistants quote. Includes an openly licensed equipment dataset, an outage runtime calculator with real validity checks, a 100-question knowledge base, and a self-hosted email capture running on Cloudflare Workers KV.
Want something built to this standard?
Tell me what you are trying to do and I will tell you honestly whether I am the right person for it — and roughly what it costs. Replies within one business day.