TerraFormer: Automated Infrastructure-as-Code with LLMs Fine-Tuned via Policy-Guided Verifier Feedback

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jana, Prithwish, Davidson, Sam, Bhasker, Bhavana, Kan, Andrey, Deoras, Anoop, Callot, Laurent
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915726531493888
author Jana, Prithwish
Davidson, Sam
Bhasker, Bhavana
Kan, Andrey
Deoras, Anoop
Callot, Laurent
author_facet Jana, Prithwish
Davidson, Sam
Bhasker, Bhavana
Kan, Andrey
Deoras, Anoop
Callot, Laurent
contents Automating Infrastructure-as-Code (IaC) is challenging, and large language models (LLMs) often produce incorrect configurations from natural language (NL). We present TerraFormer, a neuro-symbolic framework for IaC generation and mutation that combines supervised fine-tuning with verifier-guided reinforcement learning, using formal verification tools to provide feedback on syntax, deployability, and policy compliance. We curate two large, high-quality NL-to-IaC datasets, TF-Gen (152k instances) and TF-Mutn (52k instances), via multi-stage verification and iterative LLM self-correction. Evaluations against 17 state-of-the-art LLMs, including ~50x larger models like Sonnet 3.7, DeepSeek-R1, and GPT-4.1, show that TerraFormer improves correctness over its base LLM by 15.94% on IaC-Eval, 11.65% on TF-Gen (Test), and 19.60% on TF-Mutn (Test). It outperforms larger models on both TF-Gen (Test) and TF-Mutn (Test), ranks third on IaC-Eval, and achieves top best-practices and security compliance.
format Preprint
id arxiv_https___arxiv_org_abs_2601_08734
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TerraFormer: Automated Infrastructure-as-Code with LLMs Fine-Tuned via Policy-Guided Verifier Feedback
Jana, Prithwish
Davidson, Sam
Bhasker, Bhavana
Kan, Andrey
Deoras, Anoop
Callot, Laurent
Software Engineering
Artificial Intelligence
D.2.4; D.2.9; F.3.1; I.2.6; I.2.7
Automating Infrastructure-as-Code (IaC) is challenging, and large language models (LLMs) often produce incorrect configurations from natural language (NL). We present TerraFormer, a neuro-symbolic framework for IaC generation and mutation that combines supervised fine-tuning with verifier-guided reinforcement learning, using formal verification tools to provide feedback on syntax, deployability, and policy compliance. We curate two large, high-quality NL-to-IaC datasets, TF-Gen (152k instances) and TF-Mutn (52k instances), via multi-stage verification and iterative LLM self-correction. Evaluations against 17 state-of-the-art LLMs, including ~50x larger models like Sonnet 3.7, DeepSeek-R1, and GPT-4.1, show that TerraFormer improves correctness over its base LLM by 15.94% on IaC-Eval, 11.65% on TF-Gen (Test), and 19.60% on TF-Mutn (Test). It outperforms larger models on both TF-Gen (Test) and TF-Mutn (Test), ranks third on IaC-Eval, and achieves top best-practices and security compliance.
title TerraFormer: Automated Infrastructure-as-Code with LLMs Fine-Tuned via Policy-Guided Verifier Feedback
topic Software Engineering
Artificial Intelligence
D.2.4; D.2.9; F.3.1; I.2.6; I.2.7
url https://arxiv.org/abs/2601.08734