Claudio - Cloud architecture design simplified using AI.

nspiration

We were tired of whiteboarding cloud diagrams that go stale the moment you hit “deploy.” Sprawling docs, tribal knowledge, and ad-hoc infra PRs made even simple changes slow. We wanted a teammate that speaks architecture, understands constraints, and outputs something you can actually run, without ten meetings and three RFCs.

What it does

Claudio turns plain-English requirements into production-grade cloud architectures. You describe workloads, SLOs, budget, and constraints; it proposes designs, renders diagrams, and emits IaC (Terraform/CDK) for AWS/Azure/GCP plus guardrails, cost callouts, and security notes. It can diff “current vs proposed,” explain trade-offs, and generate migration plans or runbooks.

How we built it

Under the hood: a retrieval + planning layer, fine-tuned LLMs (Llama 3.x & Gemma family) with an architecture schema, and a compiler that maps design choices to reusable IaC blueprints. We wrapped it in a React/Tailwind front end with a live diagram canvas, and a Python/FastAPI backend that orchestrates the graph (policy checks, cost estimators, provider SDK lookups) before emitting validated Terraform/CDK.

Challenges we ran into

Cloud nuance is gnarly: service defaults, regional quirks, and “works in prod except Tuesdays.” Keeping outputs deterministic while staying flexible was tricky. Cost and security trade-offs needed clear, explainable reasoning, no black box. And cross-cloud abstractions… yeah, naming things is still the hardest problem.

Accomplishments that we're proud of

Claudio can: generate a multi-region, zero-trust, autoscaled stack from a paragraph; produce runnable IaC with least-privilege IAM; and attach a clean diagram + diff view you can drop in a PR. Seeing teams go from napkin sketch to mergeable infra in under an hour felt like cheating (in a good way).

What we learned

Great DX is as important as model quality. Guardrails, validations, and “why this, not that” explanations build trust. Also, cost, reliability, and simplicity pull against each other, surfacing those tensions early saves a ton of rework.

What's next for Claudio - Cloud architecture design simplified using AI.

Short-term: deeper drift detection, inline threat modeling, and live cost envelopes during design. Mid-term: environment promotion flows (dev→stage→prod) with policy packs, and richer multi-tenant patterns. Long-term: closed-loop ops, Claudio proposes, deploys, observes, and iterates safely, so your architecture stays healthy as reality changes.

Thinking ChatGPT can make mistakes. Check important info.

Built With

• amazon-web-services
• fastapi
• gemma
• rag
• redis
• sagemaker
• typescript