Skip to content

WHY.md

Latest commit

 

History

History
154 lines (108 loc) · 6.27 KB

WHY.md

File metadata and controls

154 lines (108 loc) · 6.27 KB

Why This Tech Stack?

English | 한국어

This document explains the reasoning behind each technology choice in juny.

Backend

FastAPI

  • AI/ML ecosystem: Direct access to Python's AI libraries (LangChain, Transformers, etc.)
  • Async-first: Built on Starlette with native async/await support
  • Auto-generated docs: OpenAPI (Swagger) and ReDoc out of the box
  • Pydantic validation: Request/response validation with type hints
  • Scalability: Easy horizontal scaling with stateless design

SQLAlchemy (async)

  • ORM flexibility: Can write raw SQL when needed, ORM when convenient
  • Async support: Native asyncio with asyncpg driver
  • Migration-friendly: Alembic integration for schema versioning
  • Mature ecosystem: Battle-tested in production for decades

PostgreSQL 16

  • ACID compliance: Data integrity guaranteed
  • JSON support: JSONB for flexible semi-structured data
  • Vector extension: pgvector for AI embeddings and similarity search
  • Performance: Advanced query planner, parallel queries, partitioning
  • Extensions: PostGIS, full-text search built-in

Redis 7

  • Sub-millisecond latency: In-memory data structure store
  • Versatile: Cache, session store, pub/sub, rate limiting
  • Persistence options: RDB snapshots or AOF for durability
  • Cluster support: Horizontal scaling when needed

MinIO

  • S3 Compatible: Drop-in replacement for AWS S3 API, seamless migration to production Cloud Storage
  • Local development: Same API as production environment, no vendor lock-in during development
  • Self-hosted: Runs locally with Docker/Podman, no external dependencies or service accounts required
  • Open source: Enterprise-grade object storage with full control over data

Mobile

Flutter 3.41.2

  • Korea eGovFrame v5: Selected as the official mobile framework by Korea e-Government Standard Framework
  • Flexible versioning: Easy to pin and upgrade Flutter/Dart versions per project
  • Hot reload: Sub-second UI iteration during development
  • Native performance: Compiled to ARM, no JavaScript bridge

Riverpod 3

  • Compile-safe: Dependencies checked at compile time
  • Testable: Easy to mock and test in isolation
  • No context required: Access state from anywhere without BuildContext
  • Code generation: Reduces boilerplate with riverpod_generator

go_router 17

  • Declarative routing: URL-based navigation like web
  • Deep linking: Works out of the box for iOS/Android
  • Type-safe: Code-generated route parameters
  • Nested navigation: Shell routes for bottom nav, tabs

Forui

  • shadcn/ui for Flutter: Consistent design language with web (shadcn/ui)
  • Customizable: Themeable components with Tailwind-like token system
  • Accessible: ARIA-equivalent semantics for mobile
  • Lightweight: No heavy dependencies, just widgets

Firebase Crashlytics

  • Real-time crash reporting: Immediate visibility into production issues
  • Breadcrumbs: User actions leading to crash
  • Stack deobfuscation: Readable stack traces for Flutter
  • Free tier: Generous limits for most apps

Fastlane

  • Automated releases: One command to build, sign, and deploy
  • Cross-platform: iOS and Android with same workflow
  • CI integration: Works seamlessly with GitHub Actions
  • Metadata management: Screenshots, descriptions, changelogs

Infrastructure

Terraform

  • Infrastructure as Code: Version-controlled, reviewable infra changes
  • Declarative: Describe desired state, let Terraform handle the rest
  • State management: Track what's deployed, plan before apply
  • Modules: Reusable, shareable infrastructure components

GCP (Cloud Run, Cloud SQL, Cloud Storage)

  • Generous free tier: $300 credit for new accounts, always-free tier for many services
  • Serverless containers: No server management, scales to zero
  • Pay-per-use: Only charged when handling requests
  • Managed database: Automated backups, HA, maintenance
  • Global CDN: Cloud CDN for static assets and API caching

GitHub Actions + Workload Identity Federation

  • Keyless deployment: No service account keys to manage or rotate
  • Native GitHub integration: Triggered on push, PR, scheduled
  • Matrix builds: Parallel testing across versions/platforms
  • Marketplace: Thousands of community actions

Developer Experience

Rust-based Toolchain

We prioritize speed across the entire development workflow by choosing Rust-based tools:

  • Biome: Linter + formatter in one tool, 100x faster than ESLint + Prettier
  • uv: Python package manager, 10-100x faster than pip/poetry
  • Ruff: Python linter + formatter, 10-100x faster than flake8/black

mise

  • Polyglot monorepo support: Node, Python, Flutter, Terraform — different ecosystems, one tool
  • Project-local versions: .mise.toml ensures consistent environments across OS during developer onboarding
  • Task runner: Replace Makefile, npm scripts, shell scripts with unified mise commands
  • Written in Rust: Instant tool switching, no startup overhead

Polyglot Monorepo

  • Single repository: API (Python), Mobile (Dart), Worker (Python), Infra (HCL) in one place
  • Bounded contexts: Each language ecosystem is scoped to its directory, preventing cross-contamination
  • Atomic changes: Frontend + backend changes in single PR
  • Unified tooling: Same mise commands across all apps

Trade-offs

Choice Trade-off Why We Accept It
FastAPI over Node.js Python-only backend Python AI/ML ecosystem (Gemini, LiveKit), async-first
Flutter over React Native Larger app size, custom rendering Korea eGovFrame v5, flexible versioning
Gemini over OpenAI Google ecosystem dependency Native multimodal live streaming, Vertex AI integration
LiveKit over Agora/Twilio Self-hostable, less managed Open-source WebRTC, fine-grained role control

Summary

This stack optimizes for:

  1. Developer velocity: Hot reload, type safety, auto-generated clients
  2. Production readiness: Managed services, serverless scaling, CI/CD
  3. Team scalability: Clear boundaries, shared tooling, documentation
  4. Long-term maintainability: Proven technologies, active communities