HO2 — "Ask-Me" AI Assistant

Hands-on portfolio project · Week 2 · Solo · module M5. Part of the ForgeMind AI — AI Productivity Essentials course.

Goal

Done when: you have a shareable AI-powered Claude Artifact that answers questions about you live, grounded in your real information — not a page of pre-written answers.

What it is

A Claude Artifact (a single-file page Claude builds for you inside a chat) that:

• hardcodes your real About / projects / skills as its grounding, and
• calls Claude at runtime with the built-in window.claude.complete(prompt) API to answer visitor questions from that material only — no API key, no server, no code you host.

You build it by prompting Claude, then click Publish to get a link you can share.

Who can actually use it live (read this — it's the honest bit)

• To build it, you must turn on Settings → Capabilities → "AI-powered artifacts" in Claude.ai (one-time).
• To use the live AI in a shared artifact, a viewer needs their own Claude account — when they ask a question they're prompted to sign in, and the usage counts against their account. It is not an anonymous public bot. So the honest framing is "anyone with a Claude account can ask it live," not "a recruiter just asks and gets an answer."

One caveat: Anthropic changed how window.claude.complete behaves on 2025-07-31. Write to the documented window.claude.complete(prompt: string): Promise<string> API (the samples do), and if live answers ever stop working, open your artifact inside claude.ai and confirm the current way to call Claude — the feature may have shifted since this was written.

What to ship

A repo with four files, plus your published artifact link:

ho2-askme-assistant/
├─ index.html      ← the artifact code, exactly as Claude built it
├─ grounding.md    ← the FACTS you gave it — the material it is allowed to say
├─ README.md       ← what it is · your published link · who can use it live
└─ .gitignore

grounding.md is what makes the work checkable: the assistant may only say what is in it.

Note: you can switch on GitHub Pages and the page will load — but the chat will not answer there. window.claude.complete() only exists inside claude.ai. Pages shows the interface; the published artifact link is the thing that actually answers. Put it at the top of your README.

Pick a problem statement

Choose one — each maps to a fully-built example in samples/ you can open and copy:

1. You are job-hunting and tired of recruiters skimming past your CV. Build an AI-powered Claude Artifact — an "Ask-Me" chat grounded in your experience, projects and skills — so anyone with a Claude account can ask "have you led a team?" or "do you know React?" and get an accurate answer from your real material. Success: a published artifact link where three sample questions answer correctly, live.

2. As a freelancer with a scattered portfolio, you want one link that sells your work. Build an AI-powered Claude Artifact that explains each project, the problem it solved and the decisions behind it, grounded only in your own write-ups. Success: a shared artifact where a Claude-using visitor can ask about any project and get the real story, not invented detail.

3. You sell a product and the same pre-sale questions clog your inbox — pricing, integrations, refunds. Build an AI-powered Claude Artifact grounded in what you actually publish, so a prospect gets an accurate answer instantly. Success: a published artifact where test questions about your plans, integrations and refund policy are answered exactly from your own material.

4. Your team keeps asking the same brand questions — what is our tone, our positioning, which claims are approved. Build an AI-powered Claude Artifact grounded in your tone of voice and your approved / banned claims list, so anyone writing copy can check the rules before they publish. Success: a shared artifact that correctly approves an approved claim, refuses a banned one, and offers the closest approved wording instead.

5. Several promotions run at once and sales, support and social all quote them from memory — one wrong end-date, or a code stacked that should not be, costs real money. Build an AI-powered Claude Artifact grounded in your live offer sheet — codes, dates, eligibility, exclusions. Success: a published artifact that quotes the real terms, refuses to invent an offer, and says plainly when something is not running.

Use it with your Claude.ai subscription

No API key needed.

1. Open the landing page index.html (or view it on GitHub Pages) to browse the five examples.
2. Open the README of the sample closest to your situation and copy its example prompt.
3. In Claude.ai, enable Settings → Capabilities → "AI-powered artifacts".
4. Paste the prompt into a new chat with your own details filled in. Claude builds your page as an Artifact — refine it in the chat until it sounds like you.
5. Click Publish on the artifact, copy the link, and paste it into your README.

How to use this repo

• index.html — landing page linking the five worked examples.
• samples/sample-01 … sample-05 — five complete reference artifacts (one per problem statement), each with its own README + example prompt.
• On the starter branch you'll find a copy-and-use template with example grounding already filled in — clone it, swap in your details, publish.

Optional — automate it with the API (advanced)

You do not need this for the course. window.claude.complete runs on the viewer's Claude account, so there's no key. If you later want a page that answers without the viewer needing a Claude account, you'd host it yourself and call the Anthropic API with a key kept server-side — a separate, paid path that is never required to finish this hands-on.

---

HO2 · Solo · ForgeMind AI Course · module M5 (Week 2)

💡 Use your Claude.ai Pro plan wisely

The Pro plan has a usage limit that resets every few hours. A few habits make it stretch — and keep a mistake from burning your whole session:

• Use the example prompt in each sample's README — it's already written and tested. Don't reinvent it.
• One clear prompt beats lots of vague back-and-forth. Say what you want, with an example, in a single message.
• Start a new chat when you switch tasks. Long chats re-read every earlier message and use up your limit faster.
• Don't paste big files over and over. Paste once, then refer back to it.
• If something works, keep it. Tweak it rather than regenerate from scratch.
• Using Claude Code or Cowork? This repo's CLAUDE.md makes Claude follow these same rules automatically, and SKILL.md is a reusable "token-wise" skill.

If you do hit the limit, it resets after a few hours — nothing you've saved is lost.