MarketReality OS

AI hallucinates crypto data. MarketReality OS stops it — every AI claim must cite real CoinMarketCap evidence or it doesn't get written. Evidence before interpretation, always.

Comment

What inspired me

I got genuinely frustrated with how AI handles crypto questions.

Ask any AI assistant "is SOL a good trade right now?" and it gives you a confident, detailed answer — with completely fabricated numbers. Wrong prices. Invented volume figures. Made-up exchange data. And it never says "I don't know." It just sounds authoritative and lies.

That bothered me enough to build something about it.

What I built

MarketReality OS is a deterministic market intelligence engine with one core rule: Evidence Before Interpretation.

Here is how it works:

  1. You enter any crypto symbol (BTC, SOL, ETH, etc.)
  2. The backend connects to the CoinMarketCap Pro API and fetches live data — real price, real volume, real exchange distribution
  3. That raw data gets converted into numbered Evidence Objects — structured, verifiable facts like [EV-001] for price, [EV-004] for exchange coverage
  4. The AI is then given only those evidence objects to work with — it cannot use anything from its training data
  5. Every sentence in the AI report must cite a specific evidence ID. No evidence = no claim.

The result is a market integrity report where every finding is 100% traceable back to real CMC data.

Modules

Module What it does
Reality Audit Live CMC data with full structural analysis
Evidence Investigator AI report fully grounded in evidence objects
Trade Reality Can this asset actually absorb your order size?
Stress Lab What happens if top exchanges go offline?
Evidence Vault Every raw evidence object, fully transparent
Market Passport Clean summary card of the full audit

How I built it

  • Backend: Python + FastAPI, deployed on Render
  • Frontend: Next.js (App Router), deployed on Vercel
  • Data: CoinMarketCap Pro API — live quotes, market pairs, global metrics
  • AI Layer: Evidence-locked report generation — AI can only reference provided evidence objects

Challenges I faced

The biggest one was the CoinMarketCap free tier blocking the /market-pairs endpoint. Instead of crashing the UI with an error, I built a Fallback Simulation Engine that detects the API paywall and seamlessly injects realistic simulated market structures. The platform stays fully functional and the user experience is never broken — it just transparently labels the data as simulated.

Deploying a Python backend and Next.js frontend as two separate services (Render + Vercel) and getting CORS to work across both was also more painful than expected.

What I learned

Building fault-tolerant systems that degrade gracefully is genuinely hard. Most tutorials teach you the happy path. Real production software needs to survive when a third-party API sends a 403, a rate limit, or just goes down entirely. That was the most valuable lesson from this project.

Try it live

https://marketreality-os.vercel.app/

Built With

Share this project:

Updates

Submission history