Helicity

AI powered liquidity stress scoring with verifiable audit trails

Hackathon · Cornell AI Hackathon 2026, Programmable Capital track · Team of 6 · Backtesting, IPFS audit trail, dashboard UI, and slides · March 2026

Overview

Helicity scores liquidity stress for stablecoin reserves. It was built with a team of 6 in the Programmable Capital track at the Cornell AI Hackathon 2026.

How it works

A NetworkX knowledge graph links stablecoins, banks, and jurisdictions. Composite risk combines reserve duration, FDIC Call Report data, and weather tail events, and a multi-model LLM jury using Claude and Gemini writes consensus causal narratives. Every score snapshot is pinned to IPFS, and a FastAPI backend plus a five tool MCP server expose scores to the React dashboard and to AI agents.

My role

I wrote 57% of the commits, across the backend, dashboard, and pitch.

IPFS pinning through Pinata, plus a trust badge that verifies it
A Hurricane Ian backtest service and timeline pages with a scrubber and per dimension breakdown
A mock Chainlink oracle endpoint shown in the trust badge
Connecting the backend framework to real data APIs and work on the stress score
The full dashboard UI overhaul, landing page, and the React slides app

Status

The dashboard and slides are live, and the repo includes one command setup for running the backend, MCP server, and frontend locally.