About

Tabkeel is built by one person: Francisco Ferreira, developer and founder, in Brazil. No invented team in the footer, no stock photos. The legitimacy front of our own exam looks for exactly that kind of staging, so here is the opposite of it: the small truth.

It exists because I build products with AI and lived the pattern the exam looks for. I shipped SaaS with a beautiful page and a blank 404. I kept, without knowing, one price on the pricing page and another in practice at checkout, and found out from the data, the expensive way. AI builds the happy path at absurd speed, and stops. Everything else, the unhappy path, cross-page coherence, machine readability, stays quietly broken until a customer finds it.

Tabkeel is the tool I wish I had pointed at my own launches: paste the URL, get the list of what is broken, contradictory or unreadable to AI, with the proof next to it, and the fix written to paste back into the agent that built the site.

The three rules

• Clickable evidence, always. A finding without a link, number or literal excerpt does not ship. If you think we got it wrong, the proof is right there to check.
• The LLM locates, it never judges. No AI gives an opinion about your site here. The verdict is deterministic code; the model only finds excerpts and drafts fixes.
• Underreport, on purpose. When in doubt, the finding dies. Better a short, indefensible list than a long, debatable one.

What else I do

I am also the author of PromptEval, an evaluator and registry for teams shipping LLM prompts to production. Both products share the same stubbornness: replace opinion with verification.

Talk to me

Direct email, no form: francisco@tabkeel.com. Found a false positive? Send the exam link and I will fix the verifier. Thank you in advance.