Y-Research-SBU/QuantHarness

What it solves

QuantHarness is a multi-agent system designed to automate technical analysis for high-frequency trading. It solves the problem of synthesizing complex market data—such as technical indicators, chart patterns, and trend lines—into actionable trade directives (LONG or SHORT) without requiring a human trader to manually interpret every chart.

How it works

The system uses a graph-based multi-agent architecture built with LangChain and LangGraph. It employs four specialized agents that process market data in parallel and then synthesize the results:

• Indicator Agent: Calculates technical metrics like RSI, MACD, and Stochastic Oscillators from raw OHLC data.
• Pattern Agent: Generates price charts and uses a vision-capable LLM to identify chart patterns (highs, lows, and shapes).
• Trend Agent: Creates annotated K-line charts with fitted trend channels to determine market direction and consolidation zones.
• Decision Agent: Acts as the final synthesizer, combining reports from the other agents to recommend entry/exit points and stop-loss thresholds.

Who it’s for

This tool is intended for quantitative traders, researchers, and developers interested in integrating LLMs and computer vision into financial market analysis.

Highlights

• Vision-Driven Analysis: Requires LLMs with image input capabilities to analyze generated price charts and trend channels.
• Cros-Asset Support: Works with stocks, cryptocurrencies, commodities, and indices via Yahoo Finance.
• Flexible Timeframes: Supports analysis from 1-minute to daily intervals.
• Multi-LLM Compatibility: Compatible with OpenAI, Anthropic (Claude), Qwen, and MiniMax models.
• Integrated Web UI: Includes a Flask-based interface for real-time asset selection and API management.