yappy.fyi · hackathon 2026 · Carter — research · Ege — product views & pipelining · Ayda · Max — Pioneer fine-tuning

Adaptive Global Campaigns

Campaign intelligence that learns how your product should work in every market: we contextualise your brand per market, decide what adapts, create, execute — and learn from the response.

Ingest→Contextualise→Plan→Produce→Execute→Observe→Interpret→Adapt↺
Campaign Model×Market Model→Adaptation Plan→Strategy + Creative

The stack

OpenAIReasoning
understands brand material; strategy, planning, interpretation
TavilySensing
live market research: competitors, trends, culture per geography
Pioneer / FastinoSpecialisation
small model that learns the brand's campaign judgement
falGeneration
market-specific visual assets and variations
VEEDProduction
selected concepts → finished video creatives
H CompanyExecution
computer-use agent runs approved operations

Tavily senses → OpenAI plans → VEED creates

Architecture v2

Ege

1 · Ingestion → Model gen

Ingestion layer takes brand artifacts (Canva, docs) + Tavily; Model gen refines via a Tavily/GenAI search loop.

2 · Market analysis — "CMO talk"

Market analysis persona backed by a Pioneer-specialised model → decide on audiences.

3 · Creative design — "CCO talk"

Creative design with its own search loop, gated by a Brand Critic before anything ships.

Cultural adaptation layer

Sits over creative generation: VEED (video) + fal (visuals) produce per-market variants.

Eval: HF + Pioneer evaluator

Human feedback and a Pioneer evaluator score outputs → only high-quality reviews pass. Max fine-tunes these foundational capabilities to feed the app engines.

4 · Growth

Final stage: approved creative flows into the growth engine.

Use case: Airbnb guest mgmt

Ege's app → Japan: keigo formality, LINE-first comms, omotenashi, Rakuten/Jalan ecosystem.

Marketing inspo

Carter

Marketing agents = new coding agents

Research → create → publish → learn, looped. Yappy is that loop, specialised on geography.

via Isenberg

The UGC formula

Never show or name the app. Pain point + solution; curiosity headline + captions only.

via ads4apps

Copy proven formats

Dissect 76M-view templates and product-swap per market. Build a format library.

via consumerxai

Volume is the moat

Winners post 30–50 videos/day, same script. We automate that × languages.

via ads4apps

Variants beat net-new

Generate from a proven seed + references (the brief), never from nothing.

via ads4apps

Agent-ready distribution

TikTok ships an official MCP for campaigns, reporting & creative ops.

via mohammedguelli1

PRD locks in

Team

The thesis

Not "what content?" or "how's it performing?" — given who this product is and what's happening in this market, what should it do next?

Invariant vs adaptive

Emotional identity & visual quality never change; messaging, imagery, channels, creators localise — with reasons.

Two kinds of feedback

Market feedback (what performed) ≠ human feedback (what felt right). Performance must never erode brand identity.

Interpretation over metrics

Never "CTR +21%" — what we learned about audience, market, brand, and what changes next. Learnings accumulate as state.

Seeded story

Social app enters Manchester: £3k, 500 users, students, intimate & spontaneous — never corporate networking.

Demo, beat by beat

  1. Show messy source artifacts
  2. Build Campaign Model
  3. Select Manchester, UK — market intel populates
  4. Universal vs adapts-here split
  5. Strategy: how we'd enter
  6. Creative: 3 locally adapted concepts
  7. Generate one asset / video
  8. Reveal seeded V1 performance
  9. Human feedback: "doesn't feel like us"
  10. AI interprets both signals
  11. Adapt Campaign → V2 visibly changes

Open questions

Naming: yappy vs geomorph — one brand fronts the hackathon.

Demo product: Manchester social app (PRD) or Ege's Airbnb app → Japan?

Real vs seeded by demo day — cached fallbacks everywhere, APIs never block.

Map repo packages (core/tavily/veed/fal) onto PRD pipelines (buildCampaignModel…).