The current public Fradium beta surface, captured 11 July 2026, shows the wallet and analysis entry points. It documents availability, not a transaction guarantee, and does not replace the login-protected verdict flow.
Outcome
Fradium's six-person team won the Fully On-Chain Track at the World Computer Hacker League 2025 Global Finale.
Competition recognition for the six-person Fradium team; not evidence of production adoption or a security certification.
Crypto users often reach a send confirmation with little context about the destination address. Wallet balances, explorer history, risk signals, and community reports are fragmented, so the most important judgment happens with incomplete evidence.
Intended users
Retail crypto users who want a pre-transaction risk check in a wallet or browser extension.
Community reporters and validators who contribute evidence about suspicious addresses.
Wallet, dApp, and exchange developers exploring an address-analysis API.
Contribution
Leader & Full-Stack Developer
Wildan led a multidisciplinary student team across product development, web, wallet, extension, AI, and Internet Computer work.
Product direction, team leadership, and delivery coordination.
Full-stack implementation across the web application and wallet touchpoints.
Integration across the browser extension and canister-backed product flows.
Risk-analysis, transaction-decision, and demonstration experience.
Decisions
Choices that shaped the build.
01
Put the risk signal inside the transaction decision
A separate analyzer is easy to ignore. Fradium repeats the same evidence at the point where a user is about to confirm or cancel a send.
The product makes risk visible without taking the final decision away from the user.
02
Compose AI and community evidence conservatively
A single opaque score would hide disagreement. An unsafe model result stops the path, while an unsafe community result can override an AI-safe result.
The verdict is explainable as a decision rule, although the model and community inputs still require careful evaluation and moderation.
03
Keep model and inference code in one deployable unit
Embedding ONNX artifacts in the Rust AI canister keeps inference code and its model version together instead of fetching a mutable model at runtime.
Deployments are easier to inspect, while updates require rebuilding and redeploying the canister.
System flow
From intent to visible evidence.
01
A user signs in with Internet Identity and enters an address or starts a send flow.
02
The client detects Bitcoin, Ethereum, Solana, or ICP and prepares chain-specific features.
03
The Rust AI canister runs its embedded ONNX model and returns a risk result.
04
A safe AI result is cross-checked against Motoko community-report data.
05
Fradium renders the verdict, confidence, analysis source, and history.
06
The user cancels or explicitly confirms the test-network transaction.
Evidence
Product and team, in frame.
Fradium's six-person team after winning the Fully On-Chain Track. Wildan led the project and worked across its full-stack system; the photograph keeps the result where it belongs: with the team that built and defended it together.
Reflection
What remains, and what comes next.
Current boundary
The case study must call the implementation a hybrid multi-chain analysis flow, not claim that every step is fully on-chain.
Bitcoin, Ethereum, and Solana wallet execution is testnet or devnet scoped.
Model metrics are experimental repository evidence, not an independent security guarantee.
Next improvements
Make disagreement between model and community signals easier to understand.
Verify native payment-link transactions on-chain before accepting completion.
Gate administrative and debug actions, and replace timestamp-derived API tokens with production-grade credential handling.
Publish reproducible model-evaluation context and a current whole-repository CI result.
Technology
React 19 / Vite 6 / TypeScript / JavaScript / Plasmo / Internet Computer / Motoko / Rust / Candid / ONNX / Python / Internet Identity
