The problem FraudScan AI solves

FraudScan AI - Automated Blockchain Invoice Verification

FraudScan AI saves auditors from 3 days down to 30 seconds to verify hundreds of blockchain invoices. Normally, they manually check Etherscan one by one—copying invoice numbers, amounts, wallet addresses (0xA→0xB format), and transaction hashes. It's exhausting and error-prone, especially since 85% of fake invoices use fabricated TX hashes that are hard to spot.

How it works simply:

1. Upload PDF/text invoice files
2. AI automatically extracts key data (invoice number, Rp amount, wallet addresses, TX hash)
3. Delivers instant risk scores:
   FULL_MATCH = 0-12% risk (Rp500M+ flagged for audit attention)
   NO_TX_FOUND = 85%+ (confirmed fraud)
   AMOUNT_MISMATCH = 60%+
   DUPLICATES = +25% risk penalty

Who uses it:

• Banks/fintech for daily Rp10 trillion+ invoice compliance
• Accounting firms processing client PDFs into audit-ready CSV reports
• Corporates catching vendor payment fraud via duplicate detection

Sample invoice format processed:
Invoice Number: INV221958
Date: 2025-01-01
Amount: Rp43,687,633
Wallet From: 0x922b99dfd97e90b535df7113a6d7cfedd101450e
Wallet To: 0xadcc72bc662b110491ad87c60619b0f36c5c8870
Transaction ID: 0x1ac7db96fe4e23f7630d5b8a4e9c7d2f1e0b3a8c9d2e5f7b1c4a6d8e0f2b3c5d
Status: APPROVED
Description: Payment for consulting services Q4 2024

Results example:
Result: NO_TX_FOUND (85%) → FAKE INVOICE!
High-value Rp675M: FULL_MATCH (12%) → Manual review needed
Duplicate invoice: DUPLICATE DETECTED (90%) → Double payment risk!

View live smart contract: [https://sepolia-blockscout.lisk.com/address/0xDc60Cd8553c970Bd85605e93C68De867f11e9d37]

FraudScan transforms blockchain invoice verification from tedious manual work into instant AI-powered audit intelligence, ready for enterprise scale on Lisk Sepolia Testnet.

Challenges we ran into

The biggest challenge was fixing the Analytics dashboard that completely broke because st.session_state.last_results was never set during analysis, causing the page to crash with KeyError when users switched from Audit to Analytics tab. The main analysis correctly stored st.session_state.results and st.session_state.last_results, but the Analytics page tried reading a non-existent key, so no charts or data appeared despite successful audits. I solved this by changing one line from results = st.session_state.last_results to results = st.session_state.results, making all three charts (Risk Distribution pie, Amount vs Risk scatter, Risk Score histogram) instantly functional.

The toughest technical challenge came when integrating the mandatory Lisk Sepolia smart contract requirement for the competition, discovered just hours before submission deadline. With zero Solidity experience and no testnet ETH, I had to learn MetaMask Lisk network setup (Chain ID 4202, RPC endpoint), hunt down working faucets.