New P4IR framework uses RL to boost LLM accuracy in code compliance systems

Researchers have developed P4IR, a novel two-stage framework designed to enhance the accuracy of large language models (LLMs) in generating automated code compliance (ACC) systems for building regulations. The framework first employs supervised fine-tuning (SFT) to imbue LLMs with domain-specific knowledge, followed by Group Relative Policy Optimization (GRPO) to refine the generated code skeletons. This approach demonstrated significant improvements, reducing tree edit distance by up to 23.8% and token-level Levenshtein distance by 38.6% compared to SFT-only baselines, while also showing a reduction in false positives. AI

IMPACT This research offers a method to improve the reliability and accuracy of LLM-generated code compliance systems, potentially reducing errors in automated regulatory checks.

RANK_REASON The cluster contains a research paper detailing a new framework for improving LLM performance on a specific task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Claude Opus
• Claude Sonnet 4.5
• GLM-4.7
• GPT-5.2
• P4IR
• Qwen-3-Max

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →