Sentinel-Mesh preprint
Research Square · manuscript under review at IEEE Transactions on Cloud Computing
Research
Research
Making AI-generated infrastructure changes verifiable: LLM remediation checked by formal methods, evaluated on an open benchmark.
Research Case Study
Sentinel-Mesh
A neuro-symbolic research framework: an LLM proposes Terraform fixes, and a Z3 SMT verifier accepts only patches that provably satisfy cloud security invariants.
- 83.81%Reported remediation rate (88/105)
- 105CloudFix-Bench cases
- 0.0%Reported security regression rate, within scope
- 29Formal proof certificates issued
Lead author and developer. Research Square preprint; the manuscript is under review. Read the research case study, including the architecture, evaluation, and limitations.
CloudFix-Bench
A benchmark of 105 AWS Terraform misconfiguration patterns across 8 infrastructure pillars and 60+ AWS service types, built to evaluate Sentinel-Mesh against a Checkov baseline, a no-witness ablation, and an external set of 12 real-world cases. Archived on Zenodo for reproducibility.
Publications & artifacts
CloudFix-Bench dataset
Open benchmark archived on Zenodo
Source code
Sentinel-Mesh framework, runners, and benchmark harness
Technical article
Beyond Heuristics: Formally Verifying AI-Generated Infrastructure with Z3 SMT Solvers
IEEE Access peer review
5 verified reviews, recorded on Web of Science and ORCID
Research profile
Publications and research activity
Hire me
Available for remote AI automation, n8n, AI agent, API integration, and B2B SaaS QA projects.