StratEngine: Quant Strategy Research Engine

Overview

StratEngine is built for fast, disciplined strategy iteration under contest constraints. It turns a plain-English brief into a typed competition context, retrieves cards from a curated quant knowledge base, generates and critiques candidate strategies via multi-agent synthesis, and writes submission-ready artifacts. A plugin layer adapts the same pipeline to multiple competition formats.

Highlights

• Multi-competition plugin support: IMC Prosperity (Python Trader submissions), WorldQuant IQC/BRAIN (DSL alpha expressions with refinement loop), QuantConnect (LEAN scaffolds)
• Typed pipeline: context_builder → retriever → generator → critic/scorer → competition plugin → memory/
• Refinement loop on tested BRAIN expressions using Sharpe / fitness / sub-Sharpe / turnover metrics
• Source-quality, prompt-performance, and retrieval-behavior tracking for closed-loop iteration
• Anthropic API + Claude Code CLI fallback, Pydantic schemas for contexts and strategy cards
• Strict data hygiene — competition datasets and generated submissions stay outside the repo