RAG-RuleSync

RAG-RuleSync is an end-to-end NLP/LLM pipeline that extracts structured Design-for-Manufacturability (DFM) rules from unstructured PDFs, manuals, and specifications. It extracts rules, assigns logic, and formats them into exact CAD-ready mathematical constraints.

📚 Documentation

For an in-depth, file-by-file breakdown of the system architecture, constraints, and pipelines, please see Project_description.md.

---

Quick Start Guide

1. Installation

Install the required packages.

pip install -r requirements.txt

2. Environment Configuration

Create a .env file in the root directory and add your Groq API key.

GROQ_API_KEY=gsk_your_actual_key_here
GROQ_MODEL=meta-llama/llama-4-scout-17b-16e-instruct

3. Running the System

Option 1: Streamlit UI (Recommended)

Launch the interactive web interface to upload PDFs and visualize the extraction real-time.

python -m streamlit run simple_streamlit_app.py

Open your browser to http://localhost:8503. Output tables are saved to the output/ directory as _FINAL_FORMATTED.csv.

Option 2: Batch Processing Script

Use the batch script to process an entire folder of PDFs concurrently. Note: Be cautious with strict API rate limits when doing bulk extractions.

python batch_extract_rules.py --input /path/to/pdfs --output output/compiled_rules

Option 3: Python API

import asyncio
from core.enhanced_rule_engine import EnhancedConfig, EnhancedRuleEngine

config = EnhancedConfig()
engine = EnhancedRuleEngine(config)

result = asyncio.run(engine.extract_rules_from_text(
    'Wall thickness must be at least 0.8mm for injection molding.',
    filename='my_document.pdf'
))

print(f"Extraction result: {result}")

---

Output Format

Data flows from raw JSON extractions (output/{pdf_name}.json) into the final structured schema handled by the DFM refinement pipeline (output/{pdf_name}_FINAL_FORMATTED.csv). The CSV contains:

• RuleCategory: E.g., Sheet Metal, Turning, Injection Molding.
• Name: Semantically generated rule name.
• Operator / ExpName: The CAD-ready mathematical operators constraints (e.g., SheetMetal.Thickness >= 0.8).
• RuleText: Verbatim text.