Cloud Weaver

Cloud Weaver logo

Inspiration

Building on AWS is incredibly powerful, but getting started is often overwhelming. For hackathon teams, startups, and solo developers, translating a raw MVP idea into a concrete cloud architecture and then figuring out the exact CloudFormation templates to deploy it is a massive bottleneck. We realized there was a gap between having a great idea and actually provisioning the infrastructure to support it. We wanted to build a bridge: a tool that lets developers design visually, refine interactively, and deploy instantly without getting bogged down in AWS documentation.

What it does

Cloud Weaver is an AI-powered visual cloud architect.

Design: Users simply type out their MVP idea (e.g., "A serverless image processing app") and select the AWS services they want to use.

Visualize: The app instantly generates a visual cloud architecture diagram using Mermaid.js.

Iterate: Don't like the design? Reject it, provide feedback, and the architecture visually updates in real-time.

Deploy: Once the user approves the visual diagram, the app instantly generates a ready-to-deploy, production-grade AWS CloudFormation YAML template that maps 1-to-1 with the diagram.

How we built it

Cloud Weaver was built entirely on a modern, 100% serverless AWS stack that supports scalability and low latency.

Frontend: We built a responsive single-page application (SPA) by utilizing React and Vite, hosted on Amazon S3 and distributed globally via Amazon Cloudfront. By using react-markdown and mermaid.js to parse the AI's output, we were able to render it into a beautiful, interactive, full-screen SVG architecture diagram.

Backend: We were able to bypass the standard REST APIs and implement an Amazon API Gateway Websocket API. As a result, this routes bidrectional real-time messages directly to our AWS Lambda engine written in Python.

AI & Tool Calling: At the core of our application is Amazon Bedrock, utilizing the powerful Anthropic's Claude Opus 4.5. Instead of relying fully on unpredictable plain-text prompting, we were able to engineer strict JSON Tool Schemas (Function Calling). The LLM is then forced to output structured data, which was specifically formatted for both Mermaid.js diagrams and CloudFormation YAML templates.

Stateful Memory: The AI's wonderful ability of multi-turn memory is all thanks to Amazon DynamoDB. A unique session ID is managed by the frontend. DynamoDB's job is to store the chat history and architecture context. Users can take advantage by endlessly iterating on their design while the AI keeps track of previous feedback and context.

Custom Validation Layer: LLMS has been known to make syntax errors. To prevent this, when an LLM generates a diagram or code, it first goes through a custom-built Python validation layer. The layer strictly enforces an approved AWS service allow-list, parses for broken Mermaid.js, and validates the CloudFormation YAML structure to prevent deployment crashes.

Challenges we ran into

None of us were cloud experts when we started. Our biggest challenge was diving headfirst into AWS and researching which services actually fit our architecture, learning the nuances of the AWS console to manage our deployments, and figuring out how to debug and pivot our infrastructure when things didn't go as planned. It was a steep learning curve, but an incredibly rewarding one.

Accomplishments that we're proud of

We are incredibly proud of the end-to-end workflow we achieved during this hackathon. Taking a user from a raw MVP idea, to an interactive visual diagram and finally to a ready-to-deploy Infrastructure-as-Code script is a massive technical leap that we built from scratch.

What we learned

We gained invaluable experience with cloud architecture. Going from zero cloud knowledge to deploying a live system taught us how to provision, configure, and connect various AWS services to build a resilient backend. Beyond just making the code work, we learned the importance of reading technical documentation, monitoring system logs, and designing infrastructure that actually scales.