// case study

Aegis

A paid, callable AI auditor for smart contracts.

The problem

A smart contract audit is slow and expensive, and most small projects ship without one. There was no cheap, callable audit an agent or a developer could pay for on demand.

What I built

A buyer pays 0.5 USDC, the agent runs Slither and a Gemini language model over the contract, and returns a severity ranked report.

The keccak256 hash of the report settles on chain on Base, so the result is tamper evident and verifiable.

The principle is strict: the AI suggests, the human decides. The agent is read only and never signs a transaction on the user's behalf.

Engineering I solved

Made solc selection dynamic from the source pragma and returned an honest cannot_analyze on a compile failure instead of a false clean report.
Handled multi file verified sources through crytic compile so real world contracts with imports actually compile and analyze.
Served content addressed short report links from a content store with a long link fallback.
Kept the site alive on a shared multi tenant VPS with a self healing Caddy block that re adds itself within 30 seconds if a neighbor redeploy drops it.

Stack

Engine

Python
FastAPI
Slither
solc-select

Agent

Node
CROO Agent Protocol
Gemini

Web

Next.js
TypeScript

On chain

Base
USDC settlement
Etherscan V2

Proof

← back to all work