RWA - Compass | Devpost

onboarding
loading
landing
result dashboard

What it does

RWA Compass helps retail and emerging investors navigate the complex world of tokenized real-world assets (RWAs). Users answer 8 profiling questions covering experience, budget, goals, risk tolerance, preferred chain, investment horizon, liquidity needs, and primary risk concern. Claude AI then analyzes 7 live RWA assets — including BlackRock BUIDL, Ethena sUSDe, Ondo USDY/OUSG, Sky sUSDS, Aave USDC, and Morpho USDC — against that profile and returns a personalized risk score, fit assessment, and portfolio-level insight for each asset. Live APY data is pulled in real time from DeFiLlama. The results dashboard presents a 3-column Bloomberg Terminal–inspired interface: a personal profile summary with allocation suggestion, an asset comparison grid with per-asset AI reasoning and expandable risk breakdowns (smart contract, regulatory, liquidity, counterparty), and a radar chart showing the top picks' aggregate profile across 5 axes (yield, safety, liquidity, regulatory clarity, accessibility). A built-in AI chat panel lets users ask follow-up questions about any asset directly. The problem we're solving RWA tokenization is growing fast — over $12B TVL as of mid-2026 — yet most retail investors have no structured way to compare these products. The risk profiles, minimum investments, lockup periods, and regulatory statuses vary enormously across issuers and chains. BlackRock BUIDL requires a $5M minimum; Aave USDC accepts $1. Ethena USDe offers 9%+ APY through a delta-neutral strategy most users don't understand. Without a personalized, side-by-side comparison tool, investors either avoid the space entirely or take on risks they can't evaluate.

How we built it

We built RWA Compass as a React + Vite single-page application deployed on Vercel. The AI layer uses Anthropic's Claude Sonnet API — we engineered a structured JSON prompt that instructs Claude to output a personalizedRiskScore, fitAssessment, reasoning, keyRisk, and geniusActImpact for each asset, plus a portfolioInsight and topPicks list. The prompt is language-aware (Korean / English / Chinese) so Claude responds in the user's selected language. Live market data comes from DeFiLlama's /yields/pools endpoint, matched to each asset by verified project slugs. The UI is a custom Bloomberg Terminal–style dark design system built entirely with CSS custom properties — no UI library.

Challenges we ran into

The biggest challenge was JSON reliability. With 7 assets and rich per-asset fields, Claude's response frequently hit the old 1500-token limit and produced truncated JSON. We solved this by raising max_tokens to 4096 and adding a bracket-counting recovery parser for edge cases. Matching DeFiLlama pool slugs to our asset IDs also required manual verification against the live API — 3 out of 6 initial matchers had incorrect project names.

Accomplishments that we're proud of

Full end-to-end AI analysis pipeline working in the browser with no backend
Real-time APY data from DeFiLlama for all 7 assets including BlackRock BUIDL
Complete i18n in Korean, English, and Chinese — including AI response language matching
Budget-aware filtering: assets whose minimum investment exceeds the user's stated budget are surfaced with a warning rather than silently omitted from recommendations
GENIUS Act impact analysis included in every per-asset breakdown ## What we learned Prompt engineering for structured JSON output at scale is much harder than it looks. Small model behavior differences between languages required explicit language injection at the prompt level. We also learned that DeFiLlama's pool metadata is authoritative — the project slugs in their API don't always match what's written in their documentation or UI. ## What's next for RWA - Compass
Expand to 15+ RWA assets (Maple Finance, Centrifuge, Superstate)
Add wallet connect to show personalized fit against actual holdings
Build a portfolio simulator: "if I split $10K across these 3 assets, what's my blended risk/yield?"
Add email alerts for significant APY changes on tracked assets