AI Agentics CE Course demo resources.

The four tools, in plain English

• Claude (claude.ai or the desktop app). Used in Demos 2 and 4. The chat tool we use most often. Free tier works for short prompts. The Pro plan unlocks longer files and the Skills system that powers the Multiple Offer Sheet.
• Codex (or Claude Code). Used in Demo 1. These are agent-style tools that can write code, build n8n workflows, and connect to your own apps. We used Codex desktop in class. Claude Code from the terminal works just as well.
• n8n (n8n.io). Used in Demo 1. A workflow automation tool that connects email, calendar, AI, and hundreds of other apps. Cloud trial is fastest. Self-hosting is free if you are technical.
• Obsidian (obsidian.md). Used in Demo 3. A free notes app that becomes the human-readable view of your AI brain.

Which demo wants what

• Demo 1 (SentriLock → Calendar). Needs Codex or Claude Code, an n8n account, your work email, and your calendar. Most complex. Budget 45–60 minutes.
• Demo 2 (Multiple Offer Sheet). Needs Claude with the multiple-offer-sheet skill installed, plus your email so Claude can pull the offer PDFs. Budget 20 minutes once setup is done.
• Demo 3 (Your AI Brain). Needs Claude Code, Obsidian, and Python 3. The full walk-through has its own field guide. Budget 60 minutes hands-on, plus a 1 to 3 day wait for ChatGPT to email your export.
• Demo 4 (AI Reputation Audit). Needs Claude Code with the BioBeacon plugin. Easiest demo. Budget 15 minutes.

What you'll need

• Codex desktop (or Claude Code in the terminal). This is the AI that will help you build the n8n workflow by talking through it.
• n8n account. Sign up at n8n.io. The cloud trial is the fastest path. After the trial, the Starter plan is about $20/month. If you're technical (or you have someone who is), you can self-host n8n on your own server for free, including on a $5/month VPS or a spare Mac mini at home. More setup, zero monthly cost.
• Your work email (Outlook, Gmail, or any IMAP inbox) connected to n8n.
• Your calendar (Outlook Calendar or Google Calendar) connected to n8n.
• A sample SentriLock confirmation email saved as a .eml file. In Outlook: drag the email to your desktop. In Gmail: open the email, three-dot menu, "Download message".

Step 1 · Sketch the workflow with AI

• Open Codex desktop (or Claude Code) in a fresh chat.
• Paste the prompt below. Read the AI's response. It will list the n8n nodes you need (Email Trigger, IF, Code, Calendar Create Event, etc.) and the logic for each one.

I want to build an n8n workflow for a real estate showing email.

Goal:
When a SentriLock "Appointment Confirmed" email arrives, create a calendar event in my Outlook or Google Calendar.

Requirements:
- Poll my inbox every hour.
- Only process confirmed appointment emails, not requested/pending emails.
- Extract the property address, showing date, start time, end time, listing agent, co-listing agent, and any agent email addresses.
- If the email body only has a known team member's name, map that name to their email address.
- Create the calendar event with the property address in the title.
- Invite the listing agent and co-listing agent by email.
- Do not invite myself if I am the calendar owner.
- Add the original email details into the calendar event description.
- If parsing fails, send me an alert email instead of creating a bad event.

Please walk me through the n8n nodes I need and the exact logic for each step.

Step 2 · Feed the AI a real sample

• In the same chat, attach the .eml file you saved.
• Paste the prompt below. The AI will tell you which fields it can reliably extract and which fields are edge cases (multi-day showings, missing agent emails, weird date formats).

Here is a sample SentriLock confirmation email. Tell me what fields can be reliably extracted and what edge cases I should protect against.

Step 3 · Get the n8n workflow itself

• If you've already connected your n8n account to Codex or Claude Code, the AI can build the workflow directly in your n8n workspace. If you haven't, the AI will give you a JSON file you can paste into n8n by going to Workflows → Import from JSON.
• Paste the prompt below.

If not already created, write the n8n Code node JavaScript that extracts the showing date, time, address, listing agent email, and co-listing agent email from this email body. Use my n8n key if attached to create the workflow in my workspace, and show it in finder once it's complete. If no n8n account is attached, give the JSON that needs to be uploaded manually.

Step 4 · Add the safeguards

• This is the difference between a demo and something you'll actually keep running. The safeguards stop the workflow from creating duplicate events, processing the wrong kind of email, or failing silently.

Help me add safeguards so this workflow ignores requested appointments, avoids duplicate calendar events, and alerts me when it cannot parse an email.

Step 5 · Test, then activate

• In n8n, open the workflow and click Execute workflow. n8n will pull recent emails and run them through the flow without activating it yet.
• Check that the right calendar event was created. If not, paste the error back into the AI chat and ask for a fix.
• When it works, flip the workflow toggle to Active. n8n will check your inbox on the schedule you set.

What you'll need

• Claude Code (or Codex) on your computer, with the multiple-offer-sheet skill installed.
• Your email account connected so Claude can pull the offer attachments. Outlook and Gmail both work.
• The offer PDFs themselves, sitting in your inbox or saved to a folder.

Step 1 · Install the skill (one time)

• Download the skill folder from the Academy Tools Library.
• Drop it into your Claude Code or Codex skills directory. The skill's README walks you through where that lives on your machine.
• Restart Claude Code so it picks up the new skill.

Step 2 · Run the prompt

• Open a fresh chat in Claude Code or Codex.
• Swap the property address in the prompt below for whatever address the offers were emailed about.
• Paste the prompt. Claude will search your inbox, download the attachments, parse them, and build the sheet.

Find my emails that have the offers for [PROPERTY ADDRESS]. Download the PDF attachments and use the /multiple-offer-sheet skill.

Step 3 · Open the result

• Claude will tell you exactly where the output folder lives (something like multiple-offer-output/ next to where the PDFs were saved).
• Inside that folder you'll find a polished HTML sheet, a PDF version, and a one-page packet for each individual offer.
• Open the HTML in your browser to preview. Send the PDF to the seller.

If something looks off

• The original PDFs are the source of truth. Always cross-check the numbers on the sheet against the actual offers before you sit down with the seller.
• If an offer is missing a term (financing type, appraisal contingency), the sheet will mark it as "missing" rather than guessing. That is on purpose.
• If the extraction missed something, tell Claude: "Buyer 3's offer also has a $5,000 seller credit, please redo the sheet." It will regenerate in seconds.

Why this one has its own guide

• Demo 3 is the most powerful of the four and also the most involved. It has its own dedicated field guide on michaelwong.ai that I update every time the export format changes or a new feature lands.
• The short version of the build: kick off your ChatGPT export (it takes 1 to 3 days), request your Claude.ai export (5 minutes), and while you wait, parse your local Claude Code history into markdown.
• Then run 10 parallel sub-agents that tag every conversation with the topics, tools, and entities it touched, install the recall skill, and watch Claude start reading your past work before it codes.

The big idea, in case you missed class

• Think of the vault as the human interface. Open Obsidian, hit Cmd+G, and you see your entire AI history as a graph. Search across years of chats with one keystroke.
• Think of the recall skill as the AI interface. When you say "let's work on my listing prep workflow" in a new Claude Code session, Claude reads the relevant past chats out of the vault before it does any work. No more re-explaining who your team is, what tools you use, or what you've already tried.
• Two interfaces, one source of truth.

What you'll need

• Claude Desktop with Cowork enabled. Free to download from claude.ai/download.
• The BioBeacon plugin, installed from the public marketplace at github.com/GumpperGroup/cowork-plugin-marketplace. Free.

Step 1 · Install BioBeacon (one time)

• In Claude Desktop, open the Cowork tab and click the + button.
• Select "Plugins", then "Add Plugins".
• On the Plugins page, switch to the "Personal" tab and click the + button.
• Paste this marketplace path: GumpperGroup/cowork-plugin-marketplace
• Click Sync, then install the biobeacon plugin from the marketplace listing.
• Restart Claude Desktop to activate the plugin.

Step 2 · Run the audit

• Open a fresh Claude Code chat.
• Swap your own name and market into the prompt below.
• Paste and run. Claude will pull your public bios from Google, score them, and write you the fixes.

/biobeacon:score-visibility

Audit [YOUR NAME], real estate agent in [YOUR MARKET].

Find my public bio/profile pages if possible. Score my AI search visibility across bio quality, platform coverage, structured data, review signals, entity consistency, and content freshness.

Then give me:
1. A simple visibility score out of 100
2. Top 3 strengths
3. Top 3 gaps
4. What to fix first
5. A rewritten AI-optimized bio
6. Platform-specific versions for website, Google Business Profile, Zillow, LinkedIn, and Instagram

Step 3 · Apply the fixes

• Take the rewritten platform-specific bios and paste them in: your website "About" page first (highest weight), then Google Business Profile, then Zillow, then LinkedIn, then Instagram.
• Re-run the audit a month later. Watch your score climb.