| # Maai — 2-minute demo script (ZOE hackathon) |
|
|
| Beats mirror the slide speaker notes. Timings assume a calm pace. |
| Assets: 3 slides + live Space + downloaded PDF as wifi backup. |
|
|
| ## SLIDE 1 — Hook (20s) |
|
|
| There's a scene in The Pitt this season that went viral. A woman arrives |
| at the ER with chest pain. Her EKG reads normal — nobody knows what's |
| wrong. Then her heart stops. It turns out the paramedics had placed the |
| EKG leads too low — they hadn't wanted to move her bra. The machine said |
| she was fine because no one looked properly. As Dr. Robby says in the |
| scene: "Women are misdiagnosed for heart attacks all the time." That |
| failure was equipment — but the pattern runs through every step of |
| women's cardiac care. Here's the data. |
|
|
| ## SLIDE 2 — Problem (25s) |
|
|
| This isn't television. Coronary heart disease kills twice as many women |
| in the UK as breast cancer. Over ten years, 8,243 women died who wouldn't |
| have, if their care had matched men's. And when a woman has a heart |
| attack, she's 50% more likely than a man to be misdiagnosed — because her |
| symptoms get dismissed as vague. That last number is the one we built for. |
|
|
| [SWITCH TO LIVE APP] |
|
|
| ## LIVE — Solution (45s) |
|
|
| This is Maai. A woman describes what she's been experiencing — her own |
| words, any language. [paste English description] Maai maps her words to |
| clinical terms a GP recognises — and look: "my husband says I'm just run |
| down" is in the record. The dismissal itself becomes evidence. Nothing |
| she says is dropped, nothing is diagnosed. |
|
|
| ## LIVE — The moment (20s) |
|
|
| [paste Chinese description] Same woman, in Mandarin. Same complete |
| record — her exact characters preserved, English clinical terms |
| alongside. And this phrase — [point at 上火] — has no Western clinical |
| equivalent. Maai doesn't drop it or fake a translation. It keeps her |
| words and flags the gap. She doesn't need English to be heard. |
|
|
| [open the PDF] This is what she takes to her appointment. |
|
|
| ## SLIDE 3 — Guardrail + close (15s) |
|
|
| One thing Maai deliberately doesn't do: conclude. No diagnosis, no |
| severity scores — if symptoms are happening right now, it says call 999. |
| The tool makes her harder to dismiss; the clinician does the medicine. |
| Next, we'd structure records around SOCRATES — the framework GPs are |
| trained on — and add advocacy coaching so she knows what she's allowed |
| to ask for. |
|
|
| ## Q&A back pocket |
|
|
| - "What if she's having a heart attack right now?" → The 999 line is on |
| screen and printed on every PDF. Maai is patterns over time, never |
| triage. |
| - "Is the 50% stat solid?" → Prof. Chris Gale's study, via BHF Bias and |
| Biology (2019). Primary links in SOURCES.md. |
| - Second source on discomfort costing lives: UPenn — bystanders less |
| likely to perform CPR on women; men 23% more likely to survive public |
| cardiac events. |
| - "Why no framework/CrewAI?" → Three pure LLM transformations, no |
| tool-use or looping. Knowing when not to add a framework is the |
| engineering choice. |
| - "Chinese accuracy?" → Verbatim is preserved untouched by design; |
| clinical mapping is an advocacy aid, the clinician interprets. |
| |