AIMM Session — November 27, 2025: Vibe Coding, Schema.org Authority, and the Sourdough Excavator

30-Second Summary

Pre-holiday session packed with practical firepower. Group swapped experiments with Claude Code and vibe coding, surfaced breakthrough on AI memory and knowledge management, dove into real-world NHS voice agent case study (71% wait time reduction), and Lou dropped biggest GEO opportunity: Python-generated, schema.org-powered FAQ page that gets you cited by ChatGPT, Perplexity, Grok without blog posts.

1. Vibe Coding Is Real — But the Learning Curve Bites

Kasimir: using Claude Code, went from zero to working Pinecone interface in under an hour.

Lou’s caveat: real workflow is talk until requirements right → generate requirements doc → feed to coder → implement one thing at a time. Skip that, get rabbitholes that eat your day.

Lesson: knowing what questions to ask is the real skill. Code is easy. Prompt is the craft.

2. The Knowledge Management Gap Nobody’s Solved

Don: ChatGPT has rich history but can’t find anything. Ends up copying last two artifacts into folder structure.

Elizabeth: Perplexity artifact view saves outputs to Google Drive without copy-pasting. She used Deep Research to find updated Infinite Prompt Generator versions, synthesize hybrid, turn into Claude skill.

Lou’s pipeline: Perplexity artifact → Google Drive → N8N trigger → Qdrant vector store.

Unsolved gap: platforms aren’t designed as repositories. Google best positioned because it’s already your file system.

3. Voice Agents in Healthcare: DORA Case Study

Bally shared UK university project: AI voice agent for post-cataract surgery follow-up at NHS.

Results: 35 weeks → 10 weeks patient follow-up / 92% patient satisfaction from elderly population / Built by startup with Oxford/Newcastle linguistics researchers

Key insight: wasn’t the AI, was the linguistics. Tracked interruptions, latency, filler words, cultural patterns.

Healthcare win because: Staff resist outbound calls. Bots never fatigue. No voicemail dead-ends. Multilingual without misinterpretation.

4. The Psychographic Hub: GEO Without Backlinks

Lou unveiled: “single most powerful authority play we’ve ever had.” Schema.org structured data in single HTML page.

What tool does: Takes ICH as input → Extracts fears, beliefs, values, desires → Generates ~400 long-tail questions ideal client is Googling at 3 AM → Answers in your voice → Embeds entity signals to Wikidata + causal chains → Wraps in schema.org markup

Result: Paste code block into HTML on site. No visible content. No backlinks. Machines do indexing.

Projected outcomes (Grok estimates): Google knowledge graph: 4-5 days / 5K-50K impressions/month within 3-6 months / 200-2000% unbranded search spike / 5-30+ AI citations/month in Perplexity/ChatGPT/Grok

Lou ran Python script (geox1.py) live. Not live yet, packaging for easier deployment.

5. The Sourdough Excavator: Monetizing What AI Can’t Replace

Lou: “Monetize Your Bias”

Sourdough analogy: what makes it valuable is introducing a contaminant. Your unusual element, weird perspective, hard-won scar tissue = your starter.

Flash Boys parallel (Michael Lewis, via Don): Mike Novogratz made 8 figures arbitraging spread. Then algorithms got faster, spread collapsed. Information didn’t disappear — gap did.

Lou’s reframe: AI converges to average. Value at edges. What AI can’t do: Say “won’t work — tried it” / Pattern recognition from 20 years watching launches fail / Know what NOT to do

Business model: give commodity away (frameworks, templates, FAQs) free/cheap. Charge premium for judgment, experience, coaching preventing expensive mistakes.