ReliefPool

ReliefPool uses real-time earthquake data and machine learning to assess tsunami risk, then uses Solana to automatically release disaster relief funds when risk is high.

Comment

ReliefPool

Description

Our project aims to help non-governmental organizations (NGOs) respond quickly to tsunami-related disasters. We use machine learning to assess tsunami risk from earthquakes and Solana's high-speed transactions to automatically release relief funds when the risk exceeds a set threshold.

Our Inspiration

Natural disasters can happen in minutes, and sometimes financial relief can take days or even weeks to reach people who need it. We were inspired by this gap between the speed of a disaster and the speed of response.

Implementation

Solana transactions

ReliefPool leverages Solana's high performance network to facilitate fast speed, low cost transactions. We picked the Anchor framework for Rust's highly efficient compiled code. Our Anchor service receives a request from the Oracle service, that is, trigger_payout when the conditions are met, and after running a rules check to ensure the vault is able to pay out the balance, fires the function to create the transaction between the vault wallet populated by donation wallets to admin specified relief group/stakeholder wallets.

Oracle

Oracle reads earthquake's magnitude and compares it with the MIN_MAGNITUDE to determine whether event is sent to the classifier. In an event the classifier breaks, or could not respond Oracle would return failed and the payout is never sent. Furthermore, to ensure that no payee recieves the money twice, the Oracle compares the ID of the earthquake to ensure each one is unique. Lastly, the Oracle compares the riskScore with the payout threshold to determine whether the trigger_payout function should be called. If successful, then the trigger_payout is called, and the event's status is changed to pending and paid after confirmation.

Machine Learning Classifier

In our Implementation of the classifier, we used USGS earthquakes dataset for earthquake magnitude and depth, and NOAA tsunami dataset for longitude, latitude, and binary target label as our training and testing set. The combination of these datasets allowed us to use logistic regression as our main model. Our final model achieved 87% accuracy, 84% recall and ROC-AUC score of 0.97.

How to run

Prerequisites

The Solana program is already deployed on devnet, so running ReliefPool locally only needs the three off-chain services.

Required

  • Git, to clone the repository.
  • Node.js 20.19+ or 22.12+ with npm, for the oracle service (backend/oracle, Node 20+) and the frontend (frontend, Vite 7 needs 20.19+ or 22.12+).
  • Python 3.10+ with pip, for the classifier (backend/classifier; tested on 3.12). Its dependencies are pandas, scikit-learn, joblib, FastAPI and uvicorn (requirements.txt). The trained model is committed, so no training step is needed.
  • Free local ports: 8000 (classifier), 3001 (oracle) and 5173 (frontend). The frontend always uses 5173, because the oracle's CORS allows only that origin by default.
  • Internet access to the USGS earthquake feed and the Solana devnet RPC (https://api.devnet.solana.com).

To use the real program on devnet (otherwise the oracle simulates the chain)

  • The oracle keypair, saved as backend/oracle/oracle-keypair.json. Only the pool's registered oracle can trigger payouts. The file is gitignored and shared privately.
  • A small amount of devnet SOL in the oracle wallet for transaction fees.

Optional

  • A Solana browser wallet (for example Phantom or Solflare) set to devnet, with devnet SOL, to try the Contribute button.
  • To rebuild or redeploy the program (backend/solana-service): Rust stable plus nightly-2025-04-15, Solana CLI 3.1.10, and Anchor CLI 0.30.1 built with Rust 1.79.

Start The Application

Clone the repository and enter the project directory. '''bash git clone [email protected]:manandrew-dev/ReliefPool.git

cd {directory} '''

Features

  • Automatic Disaster Relief Payout
  • Machine Learning Tsunami Risk Classification
  • Real Time Earthquake Monitoring
  • On-chain Relief Pool
  • Automatic Fund Distribution
  • Transparent Contribution and Payouts
  • Duplicate Payout Protection
  • Fail Safe Oracle
  • Historical Earthquake Replay
  • Web Dashboard
  • Solana Wallet Integration

Usage

  • Create a Relief Pool
  • Fund the Pool
  • Monitor earthquakes
  • Assess Tsunami Risk
  • Trigger Relief
  • Distribute funds

What We Learned

We learned how to integrate machine learning, real-time earthquake data, Solana, and a web application into an end-to-end system. We also gained experience working with imbalanced datasets, evaluating ML models beyond accuracy, designing APIs between independent components, and collaborating under a tight deadline.

What Challenges You Faced

Our biggest challenges were combining USGS and NOAA data into a reliable training dataset, handling class imbalance in our classifier, and integrating the frontend, Oracle, classifier, and Solana program under a tight deadline. We also had to balance tsunami detection against false alarms when selecting our payout threshold.

Share this project:

Updates

Submission history