AtlasGen: Your intelligent travel companion

Inspiration

Every year in Barranquilla, Colombia, when an elderly cumbiambero passes away, centuries of oral tradition die with them. The Carnaval de Barranquilla—a UNESCO Masterpiece of Intangible Heritage—faces an invisible crisis: its living memory is disappearing.

I grew up in Barranquilla watching the Marimondas dance and hearing my grandmother's stories about the Congos and their African roots. But I realized something troubling:

40% of tradition bearers are over 65 years old
Zero audio guides exist for the Carnival (unlike Rio or Venice)
Only 30% of global tourism is digitized—the rest happens offline, unrecorded

When I saw that Hugo Tour Guide won the ElevenLabs Hackathon with a generic AI travel companion, I asked myself: What if we could go deeper? What if AI could preserve not just information, but actual voices and personalities of cultural guardians?

That's when AtlasGen was born—not as another tour guide, but as a time capsule for intangible heritage.

What it does

AtlasGen: Voces del Carnaval is a conversational AI platform that lets tourists experience the Carnaval de Barranquilla through the cloned voices of real tradition bearers.

For Tourists 🧳

Talk to cultural characters: Select a Marimonda, Congo, Cumbiambero, or Garabato—each voiced by a real master
Ask anything: "What does the Marimonda mask mean?" → Get an answer in the authentic voice of a local elder
Navigate the chaos: Real-time schedule of 38+ events for Carnaval 2026 (Feb 14-17)
Understand context: Point at a costume, ask "What is this?"—the AI explains the centuries of history behind it

For Tradition Bearers 🎭

Immortalize your voice: Record 5-10 minutes → Your voice lives forever
Earn fairly: 85% of revenue goes directly to you (vs. 50-70% on competitor platforms)
Scale infinitely: Your clone can guide 1,000 tourists simultaneously while you rest

Key Differentiator

Feature	Generic AI Guides	AtlasGen
Voice	TTS / Synthetic	Real cloned voices of masters
Content	Wikipedia/LLM	First-hand oral histories
Impact	Tourism only	Cultural preservation
Economics	Platform takes 30-50%	85% to creators

How we built it

Tech Stack

┌─────────────────────────────────────────────────────────┐
│                    ATLASGEN ARCHITECTURE                 │
├─────────────────────────────────────────────────────────┤
│  Frontend        │  React + TypeScript + Tailwind CSS   │
│  Voice AI        │  ElevenLabs Conversational AI        │
│  Voice Cloning   │  ElevenLabs Voice Lab                │
│  LLM Backend     │  Google Gemini 2.0 Flash             │
│  Hosting         │  Vercel                              │
│  Data            │  JSON (38 events from official       │
│                  │  Carnaval 2026 programming)          │
└─────────────────────────────────────────────────────────┘

Development Process

Hour 1: Foundation

Scaffolded React app with Tailwind
Implemented ElevenLabs ConversationalAI widget
Created landing page with official Carnaval 2026 branding (yellow #F7B731, red #E31837, green #00A651)

Hour 2: Voice Differentiation

Cloned a sample voice using ElevenLabs Voice Lab
Built character selector (Marimonda, Congo, Cumbiambero, Garabato)
Created JSON knowledge base with oral histories and event data

Hour 3: Polish & Demo

Added event chronogram with real Carnaval 2026 dates
Implemented "Tradition Bearer Portal" mockup
Recorded demo video

Key Technical Decisions

ElevenLabs over alternatives: Best-in-class voice quality + native voice cloning = perfect for cultural preservation
Static JSON over database: Hackathon speed—38 events don't need PostgreSQL
Gemini 2.0 Flash: Fast reasoning for real-time conversation context

Challenges we ran into

1. Voice Cloning Ethics

Challenge: How do you clone someone's voice respectfully?

Solution: We designed a consent-first framework:

Explicit opt-in from tradition bearers
Revenue sharing (85%) as compensation
"Digital will" concept—the voice belongs to the family after death

2. Balancing Authenticity vs. Scalability

Challenge: Real oral histories are messy, non-linear, and full of local slang.

Solution: We created a hybrid approach:

Core knowledge comes from recorded interviews (authentic)
LLM fills gaps and handles unexpected questions (scalable)
System prompt enforces local expressions: "¡Ajá!", "¡Eche!", "compadre"

3. Competing with Hugo Tour Guide

Challenge: A very similar project won the ElevenLabs Hackathon just weeks ago.

Solution: We pivoted our positioning:

Hugo tells you what Wikipedia knows
AtlasGen tells you what only grandmothers remember

4. Data Extraction

Challenge: Official Carnaval 2026 programming was only available as images.

Solution: Manually transcribed 38 events from official graphics into structured JSON—tedious but necessary for accuracy.

Accomplishments that we're proud of

🏆 Cultural Impact

First AI system designed to preserve intangible heritage through voice cloning
Partnership framework ready for UNESCO-recognized events worldwide

🎯 Technical Achievement

Functional voice agent with character switching in under 3 hours
Real Carnaval 2026 data (38 events, accurate to official programming)

💡 Business Innovation

85/15 revenue split (creator/platform) vs. industry standard 50-70/30-50
Scalable model: one voice clone serves unlimited tourists

🎨 Design Fidelity

UI matches official Carnaval 2026 branding (extracted from real promotional materials)
Festive, accessible interface for all ages

📊 Validation

Carnaval de Barranquilla attracts 2+ million visitors annually
Average tourist spends 4 days navigating without guides
Market opportunity: $50M+ in audio guide potential for this single event

What we learned

Technical Lessons

Voice cloning is mature: ElevenLabs needs only 5-10 minutes of audio for high-quality clones
Vibe coding works: Tools like Bolt.new and Claude can scaffold production-ready apps in hours
JSON > Database for MVPs: When data is static, skip the infrastructure complexity

Business Lessons

Differentiation matters more than features: Hugo has more features; we have a better story
Fair economics attract creators: 85% revenue share is a moat, not a cost
Cultural events are underserved: Rio, Venice, New Orleans—all lack AI-native guides

Personal Lessons

Heritage is fragile: Talking to elders about their craft is emotional and urgent
AI can amplify, not replace: The goal isn't to eliminate human guides, but to extend their reach
Constraints breed creativity: 3 hours forced brutal prioritization

What's next for AtlasGen: Your intelligent travel companion

Immediate (Q1 2026)

[ ] Pilot at Carnaval 2026 (February 14-17)
[ ] Record 5-10 real tradition bearers in Barranquilla
[ ] Partner with Carnaval S.A. (official organizers)
[ ] Launch on iOS/Android with offline support

Short-term (2026)

[ ] Expand to Día de Muertos (Mexico) and Inti Raymi (Peru)
[ ] Implement blockchain micropayments for voice usage tracking
[ ] Add AR mode: Point camera at costume → Voice explains history

Long-term Vision

[ ] Become the "Spotify for cultural voices"—a marketplace where tradition bearers worldwide can monetize their knowledge
[ ] Partner with UNESCO to systematically preserve endangered intangible heritage
[ ] Train AI to detect when knowledge is at risk of extinction (e.g., last speaker of a dialect)

The Dream

"In 50 years, a child in Tokyo will ask their phone about the Marimonda. And they'll hear the voice of Don Pedro, a master craftsman who died in 2030 but whose stories live forever."

That's what we're building.