TextToCode (Natural Language Query to code)

This project is an AI-powered pipeline that converts Natural Language (English) trading ideas into executable Python Trading Strategies. It allows users to describe a strategy in plain English, generates the underlying code, and automatically runs a backtest simulation on synthetic market data.

Project Overview

The system follows a multi-stage compilation process to ensure accuracy and safety:

1. NL to JSON: Uses an LLM (Gemini) to parse the English query into a structured JSON format.
2. JSON to DSL: Converts the JSON into a Domain Specific Language (DSL).
3. DSL to Parse Tree: Uses the Lark parser to validate the grammar and build a parse tree.
4. Tree to AST: Transforms the parse tree into an Abstract Syntax Tree (AST).
5. AST to Code: Compiles the AST into a fully functional Python script (simulator.py).
6. Simulation: The generated script runs a backtest engine on OHLCV data and produces a performance report.

Setup & Requirements

Ensure you have Python installed and the required dependencies:

pip install pandas numpy google-genai lark

Note: You will need a valid API Key for Google Gemini in NL_to_Json.py or configured in your environment.

How to Run

1. Run the Pipeline: Execute the main script to start the interactive converter:

   python3 run.py

2. Enter your Strategy: When prompted, type your trading strategy in plain English (see examples below).

3. View Results:

   • The system will generate a file named simulator.py.
   • It will automatically execute simulator.py.
   • Final results (Total Return, Max Drawdown, Trades) will be printed to the console.
   • A detailed trade log is saved to report.csv.

Example Prompts (Try These!)

Since the project currently uses Synthetic Data (a simulated 30-day uptrend where price moves from 100 to 165), use these specific prompts to ensure you see active trades and profit taking:

1. The Profit Taker (Simple Price Target) Captures the trend and exits at a specific price level.

"Buy when close is above 102. Sell when close is above 150."

2. Moving Average with Profit Target Enters based on a trend indicator and exits for profit.

"Buy when close is above sma of last 5 days. Sell when close is above 145."

3. RSI Overbought Strategy Buys early in the trend and exits when the momentum indicator gets too high.

"Buy when close is above 105. Sell when rsi of last 14 days is above 70."

File Structure

• run.py: Main entry point. Coordinates the entire pipeline.
• NL_to_Json.py: Handles the AI prompt engineering.
• json_to_DSL.py: Formatter for the Domain Specific Language.
• DSL_to_LarkTree.py: Defines the grammar and runs the parser.
• LarkTree_to_AST.py: Logic transformer.
• AST_to_Code.py: Code generator and Backtest Engine template.
• synthetic_OHLCV.csv: Sample market data for testing.