SafeTrade AI - Crypto Risk Analyzer

Detects crypto scams before you invest. Our Chrome extension utilizes Gemini AI to analyze smart contracts, liquidity, and ownership, providing anyone with the tools to identify rug pulls on Uniswap.

Comment

Inspiration

I’ve watched too many non-technical friends lose money to crypto rug pulls—tokens that appear legitimate on Uniswap but conceal red flags in their smart contracts. The problem? 95% of investors can’t read Solidity code or interpret on-chain data. They rely on hype, FOMO, and hope.

I wanted to build something that levels the playing field—a tool that gives anyone, regardless of technical skill, the ability to spot scam tokens before investing.

What it does

Crypto Risk Analyzer is a Chrome extension that analyzes any token on Uniswap in under 10 seconds:

  1. One-click analysis from any Uniswap token page
  2. Real-time blockchain data pulled from Etherscan and DexScreener (liquidity, holder distribution, contract verification)
  3. AI-powered semantic analysis using Google’s Gemini AI to read smart contract source code and detect malicious patterns
  4. Risk score (0-100) with clear explanations of red flags (unverified contracts, centralized ownership, suspicious functions like mint/blacklist)
  5. Side-by-side comparison of risky vs. safe tokens to educate users

The extension runs entirely in the browser—no signup, no API keys required from users.

How we built it

Tech Stack:

  • Frontend: Chrome Extension (Manifest V3), vanilla JavaScript
  • Backend: Node.js/Express API
  • AI: Google Gemini 3 Flash (for semantic contract analysis)
  • Blockchain APIs: Etherscan V2 (contract source code, holder data), DexScreener (liquidity metrics)
  • Scoring Engine: Custom deterministic risk algorithm (liquidity thresholds, age analysis, ownership checks) fused with Gemini AI risk assessment

Architecture:

  1. Extension extracts contract address from Uniswap page
  2. Backend fetches on-chain data in parallel (contract source, liquidity, holders, ownership)
  3. The heuristic scoring engine calculates base risk from hard signals.
  4. Gemini AI analyzes contract source code for semantic patterns (rug pull risks, admin privileges)
  5. The fusion algorithm combines deterministic + AI scores for the final risk assessment.
  6. Results cached per contract for demo stability

Key innovation: Context-aware scoring that distinguishes between legitimate admin features (stablecoins like USDC) vs. exploit risks (random meme coins).

Challenges we ran into

  1. Etherscan API Migration (V1 → V2)

The old V1 endpoint stopped working mid-development. Had to migrate all contract/holder queries to V2, which requires chainid parameters and has different error handling.

  1. Gemini AI Response Consistency

Initial Gemini responses were inconsistent—sometimes JSON, sometimes prose, sometimes hallucinated risk scores. Solution: Engineered a robust prompt with strict JSON schema requirements and added fallback parsing.

  1. Stablecoin False Positives

Early versions flagged USDC/USDT as “HIGH RISK” because they have mint functions and blacklists. Fixed by implementing context-aware scoring that detects stablecoin patterns (high liquidity + admin features = compliance, not exploit).

  1. Rate Limiting & Caching

Free-tier APIs hit rate limits during testing. Implemented in-memory caching (per contract address) and deterministic scoring so demos remain stable.

  1. Content Script Extraction

Uniswap’s DOM structure isn’t consistent—contract addresses appear in URLs, page text, and meta tags. Built a multi-strategy extraction system with 6 fallback methods.

Accomplishments that we’re proud of

We achieved sub-10-second analysis from blockchain fetch to scoring. Our testing revealed zero false negatives; every rugpull token was correctly flagged. The extension has a production-ready interface with clear risk visualization and is mobile-friendly. Educational demos illustrate not just which tokens are risky, but why. Integrating Gemini AI allowed us to reliably analyze contracts using prompt engineering.

What we learned

  • Blockchain APIs are fragile—always build fallbacks and don’t trust a single data source.
  • AI needs constraints—free-form LLM outputs are unreliable; strict JSON schemas + validation are essential.
  • Context matters in scoring—what’s a red flag for a meme coin is normal for a stablecoin.
  • UX is everything—non-technical users need visual clarity (risk scores, color coding) over jargon.
  • Caching saves hackathons—deterministic results prevent embarrassing demo failures.

What’s next for Crypto Risk Analyzer

Short-term:

  • Multi-chain support (Polygon, BSC, Arbitrum)
  • Historical scam database (flag tokens similar to past rugpulls)
  • Browser notifications (alert users if they’re viewing a high-risk token)
  • Export reports (PDF risk assessments for sharing)

Long-term:

  • Social features (community-flagged tokens, reputation scores)
  • Mobile app (iOS/Android with wallet integration)
  • DeFi integration (analyze LP positions, staking contracts)
  • API for developers (integrate risk scoring into wallets, DEX aggregators)

The goal: Make crypto safer for the 95% of people who can’t read smart contracts.

Built With

  • gemini-ai
  • node.js
  • chrome-extensions
  • etherscan-api
  • dexscreener
  • express
  • javascript

Built With

Share this project:

Updates