NL-to-DSL-Converter

Author: Suryansh Maurya
Contact: (LinkedIn) https://www.linkedin.com/in/suryansh-maurya — Small project built for Rootaly internship take-home.

TL;DR — what I built

I implemented a lightweight pipeline that converts plain English trading rules into a small DSL, parses that DSL into an AST, generates pandas signals (entry/exit), and runs a simple single-position backtest. This repo contains the converter, parser, codegen and a demo using synthetic OHLCV data so you can reproduce the flow in one command.

Why I built it this way

I chose a deterministic, regex-based NL→JSON mapper (no LLM) to keep the output repeatable and easy to grade. The DSL is intentionally small and explicit — easy to extend. I prioritized clarity and reproducibility over chasing exotic features.

What's in the repo

nl_to_json.py — NL → structured JSON rules (simple regex heuristics).
dsl_canonicalizer.py — JSON → canonical DSL text.
dsl_parser.py — DSL → AST (regex parser).
ast_to_code.py — AST → pandas boolean signals; includes SMA and RSI implementations.
backtester.py — simple single-position backtester (entry/exit at close).
demo.py — end-to-end demo that prints JSON, DSL, AST, signals and a backtest summary.
dsl_spec.md — DSL grammar & examples.
DEV_NOTES.md — design choices, limitations, and how I tested the project.
requirements.txt

How to run (exact steps)

python -m venv venv
.\venv\Scripts\activate
pip install -r requirements.txt
python demo.py