mdfy — a16z speedrun pitch deck (content)

8 slides. Speedrun-fit (consumer/creator + AI, 1-person founder narrative). Each slide = HEADLINE / BODY / VISUAL DIRECTION / SPEAKER NOTE. Hand to Claude as an artifact spec.

Design system: warm zinc + orange (mdfy palette). Mono captions. No emoji. Lots of whitespace. ~16:9.

Slide 1 — Title

HEADLINE (huge, top-left or center)

mdfy**The memory layer for the AI era.

SUB (small, bottom)

Raymind.AI · Hyunsang Cho · hi@raymind.ai · mdfy.app

VISUAL

  • Negative space dominant. Just the wordmark "mdfy" in display weight + tagline below.
  • Background: dark (matches mdfy.app dark theme).
  • Bottom strip: founder + URL + email in mono caption.

SPEAKER NOTE

"mdfy is the memory layer for the AI era. The way Stripe was payments infrastructure for the internet — mdfy is memory infrastructure for everyone who lives across multiple AIs."

Slide 2 — The Problem

HEADLINE

AI users live across five tools. Their memory lives nowhere.

BODY (3 short beats)

  • Every conversation starts from zero.
  • Every tool re-uploads the same context.
  • Every team re-explains its work.

VISUAL

  • 5 logo lockup at top: Claude · ChatGPT · Gemini · Cursor · MCP hosts
  • Below each, a small loop arrow pointing back to itself (every chat = isolated memory island)
  • Center bottom: greyed-out word "MEMORY" with a strikethrough

SPEAKER NOTE

"Look at how an actual AI power user works in 2026. They're in Claude for writing, ChatGPT for research, Cursor for code, Gemini for vision, MCP hosts for tools. Five contexts. Five memories. Zero shared. Every time you switch, you're paying the re-explain tax."

Slide 3 — The Solution

HEADLINE

One URL. Every AI reads it. Your knowledge, deployable like code.

BODY

  • Write in mdfy → get a URL like mdfy.app/yourhub or mdfy.app/yourbundle
  • Paste that URL into any AI chat → the AI fetches the markdown payload
  • Three layers: Doc (single) → Bundle (curated set) → Hub (your whole knowledge)
  • A concept graph ships inside each URL so the AI gets pre-built navigation, not raw text

VISUAL

  • Big horizontal arrow: [markdown editor] → mdfy.app/xyz → [Claude / ChatGPT / Gemini / Cursor]
  • The URL in the middle is the hero — show as a real-looking address
  • Doc → Bundle → Hub as nested circles or stacked layers on the right

SPEAKER NOTE

"The product is one move. You write markdown. mdfy gives you a URL. You paste that URL into any AI — it fetches, it understands, it has your context. No upload. No re-explaining. We bake a concept graph into the response so the receiving AI gets navigation, not just text. Three composition layers: a doc, a curated bundle, or your whole hub."

Slide 4 — Why Now

HEADLINE

AI memory is 2026's biggest unsolved problem. Every vendor is solving it for themselves.

BODY

  • ChatGPT Memory, Claude Projects, Gemini's saved context — all locked per-platform
  • MCP standardized tool access; nothing standardized knowledge access
  • Andrej Karpathy named the shape in his "LLM Wiki" tweets — nobody's shipping it yet
  • The cross-AI surface is open

VISUAL

  • 4 columns top: ChatGPT Memory 🔒 · Claude Projects 🔒 · Gemini Saved 🔒 · Cursor Rules 🔒 (use lock icon, not emoji — mono caption: LOCKED IN)
  • Bottom: one big horizontal bar labeled mdfy URLs — spanning all four, marked CROSS-AI

SPEAKER NOTE

"OpenAI, Anthropic, Google — they're all building memory. But they're building it for themselves. ChatGPT Memory doesn't help your Claude session. Claude Projects don't show up in Cursor. The cross-AI layer is wide open because no single AI company can fill it without canceling their own moat. That's the gap."

Slide 5 — Moat

HEADLINE

Cross-AI delivery is the moat. Any single AI company is structurally incentivized not to build this.

BODY

  • OpenAI won't ship "fetchable by Anthropic" memory. Same for Anthropic, Google, etc.
  • We sit in the gap they can't cross.
  • Distribution: every shared URL carries a viral badge → compounds.
  • Network effects: each new hub/bundle makes mdfy's collective concept graph richer.
  • The graph ships in the URL payload, not behind an auth wall → any AI inherits it.

VISUAL

  • Two-column visual: LEFT = boxed silos (OpenAI / Anthropic / Google) — each marked "won't fetch the others"; RIGHT = mdfy logo as a bridge connecting all
  • Bottom strip: 3 small icons: 🔗 viral badge | 📈 compounding distribution | 🧠 network-effect concept graph (or use Wand2/Network/Sparkles from Lucide — no emoji)

SPEAKER NOTE

"The single sentence I want you to remember: any single AI company is structurally incentivized not to build this themselves. OpenAI shipping cross-AI memory would mean shipping a feature that helps Claude. That's not a competitive moat — that's a competitive own-goal. So the cross-AI layer can only be built by an independent. That's us."

Slide 6 — Founder

HEADLINE

Designer-founder. 15 years shipping consumer software at scale. Just left a sovereign-AI lead role to do this.

BODY

  • Humain (2025–2026, Riyadh) — Product & Experience Lead, Saudi Arabia's national AI program
  • Devsisters Group Director (2021–2024, Seoul) — CookieRun VR · Meta partnership · Quest Store
  • Kakao IX UK Regional Director (2018–2021, London) — opened Kakao's first European office
  • Saudi Aramco UX Lead (2017–2018) — UX standards pre-IPO 2019, industrial VR
  • AKQA Senior UX Architect (2015–2017) — Nike (NRC, NTC), Volvo, Barclays, Virgin Atlantic
  • Education — Goldsmiths MFA + Central Saint Martins 1st Class Honours

VISUAL

  • Horizontal career timeline 2002 → 2025 with logo lockups at key roles (Humain · Devsisters · Kakao · Aramco · AKQA)
  • Bottom right: large founder photo (real) or wordmark
  • Top right: bold quote — "One Person. AI Co-founder. Infinite Output." with Raymind.AI byline

SPEAKER NOTE

"I'm not a first-time founder pivoting from a side project. I just walked out of leading product and experience at Humain — Saudi Arabia's national AI program, set the product org up from scratch. Before that I shipped CookieRun VR with Meta at Devsisters, opened Kakao's first European office, built UX standards for Aramco before the world's biggest IPO. Designer-founder. Goldsmiths MFA, Central Saint Martins. mdfy is what I'm choosing to do next — not what I'm doing because nothing else worked."

Slide 7 — Traction & Velocity

HEADLINE

Five products live in 2025. mdfy's full stack shipped in 12 weeks.

BODY (two columns)

Raymind.AI products live (2025)

mdfy.app surfaces shipped in 12 weeks

  • Web app (mdfy.app)
  • MCP server (npm: mdfy-mcp)
  • VS Code extension (Marketplace)
  • Desktop app (signed/notarized DMG)
  • Chrome extension (Web Store)
  • CLI (npm)

Numbers (fill from actual)

  • Live since: [date]
  • Hubs published: [count]
  • MCP downloads / month: [count]
  • Chrome installs: [count]
  • Build-in-public following: [count]

VISUAL

  • Two-column layout — five Raymind product logos on left, six mdfy surface icons on right
  • Bottom: three big numbers in display weight (hubs / MCP / Chrome) with mono captions underneath

SPEAKER NOTE

"This isn't a vision deck. mdfy is live. So are four other Raymind products. The thesis 'one person plus AI ships at platform speed' isn't a claim — it's a five-product portfolio. mdfy's full ecosystem — web, MCP, VS Code, desktop, Chrome, CLI — shipped in twelve weeks. Public launch is end-August. That's six weeks from today."

Slide 8 — Ask

HEADLINE

speedrun seat. Distribution muscle. Two great hires post-launch.

BODY

  • Joining speedrun: access to a16z partner network + the program's distribution / creator-economy network at the exact moment we're staking the consumer surface
  • Capital sized to: keep mdfy free during beta (no monetization friction), public-launch marketing, post-launch hires (1 generalist engineer + 1 designer)
  • Decision window: post-launch growth phase begins September 2026 — speedrun timing aligned

Demo access

Log in at mdfy.app with demo@mdfy.app to walk through a real seeded account (hub, bundles, concept index, MCP).

Contact

Hyunsang Cho · hi@raymind.ai · raymind.ai · hyunsangcho.com

VISUAL

  • Top: large headline
  • Middle: 3-row breakdown of what speedrun specifically unlocks (network / capital / timing)
  • Bottom right: prominent boxed callout — Try it now: mdfy.app · demo@mdfy.app
  • Footer line: contact info in mono

SPEAKER NOTE

"Two things make speedrun specifically the right round, not generic seed. One: your network. We're staking the consumer surface of AI memory and speedrun's creator-economy distribution is built for exactly that ignition. Two: timing. Public launch is end-August, growth phase starts September — your program window aligns. Capital sizes to keep mdfy free through beta, fund launch, and bring on two great hires post-launch. Try the product right now: mdfy.app, log in as demo@mdfy.app."

Optional appendix slides

  • Appendix A — Product screenshots: Hub view, Bundle canvas with concept graph, raw .md payload showing graph-baked-in, Galaxy constellation
  • Appendix B — Karpathy "LLM Wiki" tweet as a screenshot + line "Named the shape. We're shipping it."
  • Appendix C — Cap table / use of funds (one-page)

Design system notes for Claude

  • Palette: dark mode dominant — backgrounds #0a0a0a (zinc-950) / #1a1a1a (zinc-900), text #fafafa / #a1a1aa, accent orange #fb923c for one element per slide max.
  • Type: display sans for headlines (e.g., Inter Tight / Geist 700), mono caption for labels (e.g., JetBrains Mono / Geist Mono 500), body 16px.
  • No emoji in slides. Use Lucide icons only (Wand2 / Network / Sparkles / Link / etc.).
  • One accent per slide — orange used sparingly so it actually directs the eye.
  • Lots of negative space. Speedrun decks are scanned in 60 seconds; sparse beats dense.
  • Wordmark: lowercase mdfy with slight letter-spacing, monoweight .app suffix when shown.

Facts

  • mdfy is a URL-based memory layer that any AI can read by pasting a link into chat
  • AI users currently work across five isolated tools (Claude, ChatGPT, Gemini, Cursor, MCP hosts) with no shared memory between them
  • Cross-AI memory delivery is structurally impossible for any single AI company to build without undermining their own competitive moat
  • Founder Hyunsang Cho shipped mdfy's full stack (web, MCP, VS Code, desktop, Chrome, CLI) in 12 weeks and has five live Raymind products
  • mdfy bakes a concept graph into each URL payload so receiving AIs get navigation and context, not just raw text