aws-cost-calculator

AWS Cost Calculator is a web tool for answering "what would this architecture cost me per month?" without first knowing instance types and node counts. Instead of a traditional pricing calculator that demands a complete architecture up front, it accepts a rough service description — traffic level, stage such as MVP versus production, service shape — and returns a recommended AWS architecture with its monthly cost in USD and KRW. Two input modes are supported. The service-based flow takes a description and proposes an architecture; the manual builder lets users drag together resources such as ECS, RDS, Redis, ALB, S3, Lambda, DynamoDB and CloudFront and price that exact stack themselves. The project's central design choice is a deterministic cost engine. Every dollar figure comes from a `CostEngine` component with one method per resource type, driven by real AWS Price List Bulk API data. Region pricing is resolved through a `PricingService` that reads from a database table and falls back to a classpath JSON snapshot. AI providers are only ever asked to explain a price that has already been computed, never to compute one. AI-backed features — plain-language explanations, cost optimization suggestions and architecture recommendations — follow a consistent pattern: try the configured provider by name (Claude API, Gemini API, or a local `codex` CLI in development) and fall back to a rule-based implementation on any failure or misconfiguration. Callers see a lower-quality answer rather than an error. Optimization suggestions are credit-gated and include rule-based tips such as Graviton adoption, RDS/Redis downsizing, and NAT Gateway or CloudFront adjustments. Calculations are stored as priced-at-save-time snapshots, so a saved calculation is re-viewable exactly as it was quoted rather than recomputed against today's rates. Results can be exported as a PDF card. The stack pairs a Next.js 16 / React 19 / TypeScript / Tailwind CSS 4 frontend with a Spring Boot / Java 21 backend and PostgreSQL 16. Authentication supports Google and Kakao OAuth2 plus local email and password, using stateless JWTs. Local setup is `docker compose up -d` for the database, `./mvnw spring-boot:run` for the API on port 8080, and `npm run dev` for the frontend on port 3000; a shell script refreshes AWS pricing snapshots and requires `curl` and `jq`. Backend tests run through Maven, and the frontend has no test framework configured — lint and build are the checks used for UI changes. The README is candid about limitations: saved calculations cannot be link-shared (export is PDF only), project rows created before accounts existed have no owner and will not appear in anyone's history, local email signup has no SMTP or SES integration so `emailVerified` is only ever set by an OAuth provider, and AI optimization line items may contain model-estimated savings that are not all re-summed server-side. The roadmap lists shareable calculation URLs with OG previews, a budget-in / architecture-out reverse recommendation, and a phase two that would connect a real AWS account read-only via a cross-account IAM role to apply the same optimization rules to actual spend. Licensed under MIT.