An independent redesign concept — “v2” — of uang.ai, the website of “The AI Factory”. It reimagines what that site could be: less telling, more proving. Every claim from the original brochure is paired with an interactive proof you can poke at: a live agent demo, a blueprint concierge, interactive case studies and an ROI simulator.

As a portfolio pitch. The original uang.ai site is a static brochure — it tells you the AI Factory delivers, but never shows it. This concept demonstrates the opposite approach: the website itself behaves like the product it sells. If uang.ai wants to adopt any of it, this page doubles as the specification.

No. This is an independent concept built by Rico as portfolio work — not affiliated with or endorsed by uang.ai. Case-study metrics reference uang.ai's public website; expanded narratives and pricing examples are concept illustrations.

Concept, design and build by Rico. If you want a website (or an agent-powered product) that proves instead of tells, start with the blueprint concierge on the homepage — it produces a shareable solution draft in about five minutes.

No — it is a scripted simulation, and it is labeled as one. It runs instantly, costs nothing and stores nothing, while mirroring the real implementation: the same trace format, tool calls and human approval gate. The live reference app (UangLite) runs a real LLM agent with tool-calling and an approve/deny gate you can actually flip.

Yes, when an LLM endpoint is configured: your interview answers are polished into narrative by an OpenAI-compatible model, scoped by a deterministic rules engine so the numbers stay honest. Without a key, the rules engine alone still produces a complete, printable blueprint — the site degrades gracefully instead of breaking.

Your answers are composed into a blueprint and encoded into the shareable URL itself. There is no database, no cookies and no analytics: nothing you type is stored server-side. The flip side: the link contains the blueprint, so don't put secrets into a demo.

Case-study metrics come from uang.ai's public website; the interactive before→after narratives around them are concept illustrations. The ROI simulator and pricing ranges are explicitly labeled directional — every assumption is stated on screen, and the math is unit-tested, but they are estimates, not quotes.

Next.js 16 (App Router, Turbopack), React 19, Tailwind CSS 4 and TypeScript, with the AI SDK for optional LLM polish of blueprints and Vitest for unit tests. There is zero database: pages are statically generated and the only API route is the blueprint polish endpoint.

Because state you don't have can't leak. The blueprint is compressed (deflate-raw) and encoded as base64url into its own shareable link, so there is no server state, no PII storage, and a shared link keeps working forever — even offline. You can print it or bookmark it; the URL is the artifact.

It targets static generation, a sub-second LCP and zero tracking scripts. Content stays readable with JavaScript disabled (the Q&A sections you're reading are native <details> elements), animations respect prefers-reduced-motion, and colors switch between light and dark themes without a flash.

Yes — Vitest unit tests cover the ROI math, diagnostic scoring, the blueprint codec (round-trip encode/decode) and scenario logic. The build fails if the math behind the claims regresses, which is very much the point of a site about proof.

Source lives on GitHub (ricothenfx/uang-v2) and deploys to Vercel, with the canonical domain uangfactory.ricothen.com. The security page documents this site's own trust posture; the Q&A you're reading documents everything else.