A Novel Approach to Generating Terraform Configurations for Google Cloud Platform
summary
Claim: a RAG system plus a custom self-querying retriever and vector DB produce a deployable single-file Terraform for Google Cloud. Implementation uses LangChain, Neo4j for relations, live-sync of GitHub Terraform registries (refresh ≈ every 5 minutes), Gemini as initial model, a linter that flags and auto-fixes lines, and modular registry ingestion. Result: Cloud Weaver outputs an architecture-first conversation, generates one downloadable TF file, and successfully completes the authors’ homework first try with far fewer hallucinations than plain ChatGPT/Gemini.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:26Project motivation
- 4:15Why Terraform for GCP
- 6:22Introducing Cloud Weaver
- 8:07Live demo: Cloud Weaver
- 14:02Cloud Weaver architecture
- 18:36Results and advantages
- 19:39Future use cases
- 23:12Open source plans
- 24:37Data flow process
- 31:27Project timeline
- 34:14Handling hallucinations and failures
- 37:05Q&A
related talks
31mAI speed to value: Google Cloud & Terraform for the enterprise
14mHow Global Discovery tribe approaches Infrastructure Changes at scale
27mCase Study: Re-Thinking Our Infrastructure Tooling
55mTerraform-ing your Cloud
40mDevops Goes Promptastic
22mTerraform Hybrid: Unifying Your Physical and Cloud Datacenters
14mMulti-Cloud Foundations with Terraform
21mAccelerating cloud transformation with Terraform: A World Bank success story