Automated Strategy Generator & Compiler

Testing thousands of strategy variations across years of tick and minute-by-minute futures data (CME Nasdaq contracts) requires immense computing speed and careful memory management.

Smartush ingests granular tick data to synthesize 1-minute and N-tick bars, dynamically aggregating them into higher-timeframe series (5-minute, 10-minute, and 60-minute bars). In quantitative research and backtesting optimization, higher-timeframe calculations are often preferable for filtering microstructure noise and accelerating parameter searches, while retaining raw tick precision whenever intra-bar execution depth is needed.

An optimized in-memory data bus shares price and indicator series across test workers without redundant copies, distributing calculations across all available CPU cores using parallel processing and fast C-libraries.

Rather than assuming a single indicator tells the whole story, the engine treats market conditions as an interconnected picture of price behavior.

Automated checkers evaluate candle momentum, volume surges, volatility breakouts, and moving average crossings across lookbacks from 1 to 30 bars. When a simulated trade is entered during a baseline test, the system captures an exact digital snapshot of every active market condition at that precise moment.

Most automated backtesters fail because they try to find the 'perfect' indicator setting to enter trades, which almost always results in curve fitting. Smartush takes the opposite approach: it keeps entry rules simple and focuses on finding Dynamic Defenses: defensive veto rules that block trades during market conditions that consistently lose money.

The clustering engine groups historical trades that shared identical technical conditions. A multi-factor scoring algorithm evaluates these cohorts, heavily favoring simple, robust rules over complex ones to prevent overfitting. High-scoring defensive rules are injected into recursive backtests, systematically eliminating losing trades while preserving profitable ones.

Moving a winning strategy from Python research into live execution has traditionally been a major bottleneck, requiring software engineers to manually rewrite algorithms for target trading platforms.

Smartush automates this entire handoff with an Abstract Syntax Tree (AST) compiler. The compiler translates Python strategy rules, multi-timeframe bar series (such as 1m base with 5m and 60m aggregations), indicator configurations, and defensive vetoes into clean, strongly typed target code ready for immediate execution.

While C# and NinjaTrader 8 served as our primary production engine for CME futures and equities execution, the AST architecture is platform-agnostic: by understanding target API constraints, modular emitters can generate native code for MetaTrader (MQL), C++, or direct broker APIs, deploying the exact same defensive logic across multiple asset classes.