Coinbase Engineer Builds Crypto Trader Based on Simulated Fruit Fly Brain

A Coinbase software engineer has developed a cryptocurrency trading system powered by a digital simulation of a fruit fly's brain. The project, dubbed "Stonkfly," uses a biological connectome to process market data and execute trades on the Coinbase platform.

Created by Alex Wormuth, the system utilizes the MaleCNS v1.0 graph, a detailed mapping of a fruit fly's neural structure consisting of 166,700 neurons and approximately 25.6 million connections. To operate the trader, Wormuth maps real-time Bitcoin price charts to the simulated fly's visual inputs, treating the candlestick charts as a 320x180 display. The simulation processes this visual data by stimulating 811 R8 color inputs and 3,335 brightness inputs, using a neural readout to trigger buy, sell, or hold orders via the Coinbase AgentKit.

Repurposing Biological Structures

The project is part of a broader exploration into how biological neural structures can be repurposed for non-biological tasks. By bridging the gap between a neuroscience simulation and a live financial API, Stonkfly tests whether the architecture of a biological brain can function as a controller for external software. Wormuth has made the project open-source on GitHub, allowing other users to run the simulation in either live or paper-trading modes.

Implications for Hardware and AI

While not intended as a professional financial tool, Stonkfly serves as a proof of concept for running large-scale biological simulations on consumer-grade hardware. The system is designed to run on machines with a recommended 16GB of RAM, demonstrating that complex connectomes can be simulated without industrial-scale computing power. This intersection of neuroscience and algorithmic trading highlights new ways to approach AI controllers, moving away from traditional machine learning models toward biologically inspired architectures.

Performance and Outlook

Despite the technical achievement, the system's efficacy as a trader remains unproven. Wormuth has explicitly noted that profitable learning has not been demonstrated within the simulation. For now, the project remains an experimental venture in synthetic biology and software integration. Observers will be watching to see if further refinements to the neural readout or the integration of reward mechanisms—such as the stimulation of dopamine neurons during profitable trades—can lead to actual market viability.