CoinYield

CoinYield is an AI-native trusted infrastructure layer for DeFi risk intelligence. It builds a risk graph of cascade dependencies across pools, protocols, assets, and yield sources, helping users understand how risks can spread through interconnected DeFi systems. The project focuses on making complex on-chain relationships easier to analyze, so users can evaluate yield opportunities with more context, transparency, and confidence.

We built CoinYield by collecting data from major public sources on DeFi hacks, incidents, exploits, and postmortems. We then structured and annotated this information into a Neo4j graph database, where each node type represents a key entity such as protocols, assets, pools, incident types, dependencies, and risk categories.

For every node type, we defined risk vectors that describe how different risks can propagate through the system. These vectors were connected to real historical incidents and mapped to specific protocols, allowing us to build directed dependency graphs that show possible cascade effects across DeFi infrastructure.

The backend is built in TypeScript and queries the Neo4j database to retrieve, analyze, and construct these dependency graphs. The risk model and related graph logic are available in our repository: https://github.com/coinyield-org/risk-model

On the AI side, we integrated Claude with a RAG-based workflow, so the assistant can query the graph database, search for additional context, and combine structured graph data with external information. This lets the demo explain not only isolated risks, but also how one protocol’s failure can affect connected assets, pools, and protocols.