RugRadar

RugRadar - AI-Powered Rug Pull Detection Platform - 14 Sep to 27 Sep 2025 (Biweekly Week 7-8) - PLDG 4

Built Node.js/Express backend with RESTful API architecture for contract analysis.
Integrated OpenAI GPT-4 Turbo via API for AI-powered vulnerability detection (Reentrancy, Access Control, Oracle Manipulation,
etc.).
Connected to BaseScan API v2 (Base Chain ID: 8453) for fetching verified contract source code, ABI, creation data, and transaction
history.
Implemented Storacha/IPFS storage via MCP REST client using JSON-RPC 2.0 protocol (tools/call method with Base64-encoded
payloads).
Designed hybrid caching strategy: static audit data (AI analysis, vulnerabilities) stored on IPFS with CID indexing; dynamic
on-chain data (balance, transactions) refreshed per request.
Created rug pull scoring algorithm with weighted vulnerability analysis, transaction pattern detection (low user diversity, high
failure rates), and dangerous function identification.
Built automated contract metadata extraction pipeline (compiler version, optimization settings, proxy pattern detection, ownership
analysis).

Implemented community review system with Storacha versioning: each new review triggers JSON re-upload, generating new CID while
preserving audit data.
Enhanced AI analysis prompts with protocol recognition logic (isWellKnownProtocol, protocolName fields) for automatic
legitimacy scoring.
Increased transaction fetch limit from 100 to 10,000 for accurate metrics on high-volume contracts.
Built transaction pattern analyzer: unique interactor tracking, failure rate calculation, activity classification
(inactive/moderate/active/very_active).
Added graceful error handling for Storacha: analysis continues without caching if MCP upload fails, with detailed logging.
Optimized MCP client CID extraction with multiple fallback paths (payload.files[fileName], payload.root, payload.cid) to
handle varied response structures.
Implemented smart rug scoring adjustments: -20 points for >1000 unique users, -15 points for >10k transactions, automatic 5/100
score for AI-detected legitimate protocols.

Smart Contract Analysis Flow:
- User enters contract address → fetches source code from BaseScan.
- GPT-4 analyzes code for vulnerabilities, contract type, and legitimacy.
- Static audit cached on Storacha via MCP REST (JSON-RPC upload).
- On-chain data (balance, transactions) fetched fresh on each request.
- Rug pull score calculated using AI analysis + transaction patterns.
Storacha Integration:
- Base64-encoded JSON uploads via MCP tools/call method.
- Handles multiple CID response formats (payload.files, payload.root, payload.cid).
- Retrieves cached audits from w3s.link IPFS gateway.
- Graceful degradation if upload fails (analysis still returns to user).
What went well:
- Successfully integrated AI + blockchain + IPFS in unified platform.
- Hybrid caching saves expensive AI calls while keeping data fresh.
- AI accurately detects major protocols (USDC scored 5/100 risk).
- Community reviews persist across sessions via Storacha versioning.
Challenges:
- MCP upload response structure varied — implemented flexible CID extraction.
- Transaction limits (initially 100) showed incorrect metrics for major tokens → increased to 10k.
- Rug scoring needed adjustment for legitimate protocols → added AI-powered detection.
- SVG text alignment issues in speedometer → fixed with textAnchor and dominantBaseline.