Chart Dungeon

An AI-powered chart-reading practice RPG where players analyze live market data, clear dungeons, collect gear, and turn their decisions into personalized learning feedback.

Comment

Judging build

OpenAI Build Week judging snapshot:

https://8e484b64.chartdungeon-buildweek.pages.dev/

This stable deployment matches the version shown in the submitted demo video and will remain unchanged during judging. No account is required for the free demo.

Live production service:

https://chartdungeon.com

Inspiration

Chart education is usually presented as either a textbook or a professional trading terminal. Both can feel intimidating to beginners. I wanted to make chart-reading practice more approachable by turning it into a dungeon RPG where charts become the battlefield, questions become encounters, and learning creates visible progression.

Chart Dungeon is strictly an educational game. It does not connect to wallets, accept deposits, execute trades, or provide investment advice.

What it does

Chart Dungeon combines live Coinbase USD market charts with a pixel-art dungeon experience.

Players can:

  • Choose a market-themed dungeon route
  • Read live charts while answering evidence-based questions
  • Explore a 3x3 dungeon and fight monsters and bosses
  • Receive explanations after each decision
  • Earn equipment, experience, gold, and card packs
  • Review boss-battle records separately
  • Use Dungeon Intelligence to discover patterns in recent boss decisions

New players can enter a limited guest dungeon without creating an account. Completing the guest dungeon awards a server-verified card pack that can be preserved through sign-in.

How we built it

The frontend uses React, TypeScript, and Vite. Supabase provides authentication, PostgreSQL persistence, Row Level Security, and Edge Functions. Cloudflare hosts the application and supports the Dungeon Intelligence analysis layer. Live market candles come from Coinbase USD markets.

Chart Dungeon existed as an early public-test project before OpenAI Build Week. During the submission period, I used Codex and GPT-5.6 to meaningfully extend its onboarding, security, and educational experience.

The new work includes:

  • A no-account guest dungeon funnel
  • Server-authoritative guest sessions and battle answers
  • A compact 3x3 dungeon layout for shorter first runs
  • A shorter first-run path from entry to the boss reward
  • A chart-visible battle layout
  • A server-verified guest card-pack claim
  • Idempotent reward handling that prevents duplicate claims
  • Privacy-conscious funnel tracking
  • Error recovery for expired questions and failed connections
  • Automated TypeScript, security, and reward-flow tests

Codex and GPT-5.6 helped me reason through the gameplay state machine, inspect security boundaries, trace database and frontend errors, improve the first-time user experience, and verify the completed flow in Chrome.

GPT-5.6 was also used to plan and edit the submitted demo video, refine its English narration, and guide the voice-over production. The final video was manually reviewed and re-edited so the product footage, narration, and English subtitles remain synchronized.

Challenges we ran into

The hardest challenge was balancing a frictionless demo with secure progression. A completely client-side guest mode is easy to manipulate, but requiring an account before the first battle creates too much friction.

I solved this by keeping guest identity temporary while making battle attempts, boss completion, and card-pack eligibility server-authoritative. Guest rewards are not permanently added to an inventory before authentication, and every reward claim is processed at most once.

Another challenge was keeping the chart connected to the learning experience. An earlier question modal covered the chart and made the game feel like a generic quiz. The updated battle layout keeps the live chart visible beside the question.

Accomplishments that we're proud of

  • Players can reach a meaningful boss reward without creating an account
  • Correct answers remain hidden until the server verifies a submission
  • Guest card packs survive refresh and authentication without allowing rerolls
  • Duplicate reward claims are blocked on the server
  • The complete guest flow works from live chart to dungeon, boss, and saved pack
  • The product combines a polished game experience with a real educational purpose
  • Funnel tracking avoids collecting emails, nicknames, user IDs, or raw answers

What we learned

A game-like interface is not enough by itself. Players need an immediate connection between the chart, their decision, the result, and the next reward.

I also learned that secure progression works best when the server owns outcomes while the client remains responsible for presentation and input. This made the system more reliable without sacrificing the speed of the guest experience.

Finally, AI-generated analysis is most useful when it identifies decision patterns and explains why they matter, instead of simply repeating a player's statistics.

What's next for Chart Dungeon

Next, I plan to:

  • Expand Dungeon Intelligence with clearer personalized explanations
  • Add an evolving educational companion based on each player's history
  • Improve chart-grounded questions using live market context
  • Introduce seasonal chart-reading challenges and rankings
  • Add more equipment progression and meaningful uses for gold
  • Continue strengthening security, privacy, and accessibility

Future roadmap: I am evaluating GPT-5.6 as an optional premium analysis layer for Dungeon Intelligence, focused on longer-term performance reviews and personalized educational feedback. GPT-5.6 is not yet integrated into the live feature.

Chart Dungeon will remain a chart-reading practice game. Future AI features will focus on education and decision review, not investment recommendations or real-money trading.

Built With

  • cloudflare-pages
  • cloudflare-workers-ai
  • codex
  • coinbase-api
  • google-gemma
  • gpt-5.6
  • postgresql
  • react
  • supabase
  • supabase-auth
  • supabase-edge-functions
  • tailwind-css
  • typescript
  • vite
Share this project:

Updates