Tech Stack (CEO approved 2026-03-03)

Tech Stack (CEO approved 2026-03-03) 
 Decision Date 
 CEO Approval: 2026-03-03
 Supersedes: Express + TypeScript backend, monolithic Next.js frontend 
 
 Backend 
 
 
 
 Component 
 Choice 
 Rationale 
 
 
 
 
 Language 
 Kotlin 
 Type-safe, JVM performance, JetBrains ecosystem 
 
 
 Framework 
 Ktor 
 Coroutine-based async, lightweight, idiomatic Kotlin 
 
 
 Database 
 PostgreSQL 16 
 Rock-solid relational DB, JSONB for flexible fields 
 
 
 ORM 
 Kotlin Exposed 
 Type-safe SQL DSL, no annotation magic 
 
 
 Build tool 
 Gradle (Kotlin DSL) 
 Standard for Kotlin/JVM projects 
 
 
 Testing 
 Kotest + Testcontainers 
 BDD-style tests, real DB for integration tests 
 
 
 Auth 
 JWT (24h expiry) + refresh tokens 
 Standard stateless auth 
 
 
 Queue 
 Kotlin coroutines + Redis 
 Async jobs: label generation, carrier calls 
 
 
 
 
 Frontend — Micro-Frontend Architecture 
 
 
 
 Component 
 Choice 
 Rationale 
 
 
 
 
 Framework 
 React 19 + TypeScript 
 Stable, large ecosystem 
 
 
 Bundler 
 Vite + Module Federation 
 MFE-ready, fast HMR 
 
 
 Architecture 
 Micro-Frontend (MFE) 
 Each feature independently deployable 
 
 
 Shared UI 
 Tailwind CSS 4 + shadcn/ui 
 Design tokens in packages/ui/ 
 
 
 State 
 Zustand per MFE 
 Lightweight, no global store coupling 
 
 
 
 MFE Structure 
 
 
 
 MFE 
 Responsibility 
 Deploy 
 
 
 
 
 shell 
 Layout, routing, auth, navigation 
 Always 
 
 
 mfe-inventory 
 Inventory management, bin locations, cycle counts 
 Independent 
 
 
 mfe-orders 
 Order management, inbound/outbound 
 Independent 
 
 
 mfe-picking 
 Picking workflows, wave management 
 Independent 
 
 
 mfe-settings 
 User management, integrations, config 
 Independent 
 
 
 mfe-ai 
 Plock AI Chat, analytics, dashboards 
 Independent 
 
 
 
 Each MFE is independently deployable — a change to mfe-picking does not require redeploying the shell or other MFEs. 
 
 AI / ML Layer 
 
 
 
 Component 
 Choice 
 Use Case 
 
 
 
 
 LLM 
 Claude API (claude-sonnet-4-5) 
 Plock AI Chat — Swedish NL queries 
 
 
 ML framework 
 Python microservice 
 Isolation from JVM, flexible ML tooling 
 
 
 Route optimisation 
 OR-Tools (Google) 
 Smart Picking — VRP/TSP solver 
 
 
 Anomaly detection 
 scikit-learn 
 Z-score + isolation forest 
 
 
 Forecasting (Phase 2) 
 XGBoost / Prophet 
 Demand forecasting 
 
 
 
 
 Infrastructure 
 
 
 
 Component 
 Choice 
 
 
 
 
 Monorepo 
 Turborepo (frontend MFEs) + Gradle (Kotlin backend) 
 
 
 Mobile (Phase 2) 
 React Native — scanner + picking app 
 
 
 Cache 
 Redis 
 
 
 Testing (frontend) 
 Vitest + Playwright 
 
 
 CI/CD 
 GitHub Actions 
 
 
 Containers 
 Docker + Docker Compose (local dev) 
 
 
 
 
 Architecture Decision Records 
 ADR-001: Kotlin over TypeScript for backend (2026-03-03) 
 Context: Previous design used Express + TypeScript. Decision to move to Kotlin + Ktor.
Rationale: JVM performance for WMS workloads, stronger type safety with Exposed ORM, coroutine-based concurrency fits async warehouse operations (carrier calls, label generation).
Status: Accepted by CEO 2026-03-03. 
 ADR-002: Micro-Frontend over monolithic Next.js (2026-03-03) 
 Context: WMS has distinct feature domains that evolve at different rates (picking vs AI vs integrations).
Rationale: Independent deployment reduces risk, enables feature teams to work in parallel, allows phased feature releases.
Trade-off: Higher initial complexity, but pays off at scale.
Status: Accepted by CEO 2026-03-03.