What had to change
New Solana launches arrive faster than a person can inspect them, and most raw activity is noise. An automated strategy therefore needs more than speed: it needs repeatable detection, explainable qualification, execution controls and hard limits for when it must do nothing.
The decisive move
The engine separates discovery, qualification, risk and execution into asynchronous stages. Yellowstone streams program activity; rolling windows identify wallet clusters; offline-mined rules and a logistic model score candidates; rug checks and portfolio gates can reject a trade before Jupiter execution is considered.
What the build covered
The repository includes event ingestion, feature and rule pipelines, runtime ML qualification, configurable strategy controls, paper/live modes, position management and the operational interfaces needed to inspect why a candidate progressed or stopped.
Why it matters
The result is an inspectable research engine rather than an opaque trading script. It demonstrates event-driven Python architecture, streaming data, applied ML, blockchain integration and risk-aware automation in one system.