Fleet management system (Tisseco)

A pragmatic and solid back-end

Tisseco collects and resells second-hand clothes in thrift stores, with a strong environmental and social mission (work reintegration programs).
The goal: a management and tracking platform for their clothing collection containers, providing full traceability, reporting, and a public interface for container lookup and issue reporting.

Goals

• Production-ready back-end with clear DDD boundaries
• Multi-entity tracking: Sites → Areas → Containers
• Handling the creation and tracking of Tours (to define the “program” a driver must follow)
• Advanced geospatial search (PostGIS)
• Reliable mass CSV imports with queueing and reporting
• QR codes for seamless field operations
• Robust deployment and monitoring

Architecture

• DDD structure: presentation/, application/, domain/, infrastructure/
• Persistence: TypeORM (migrations, seeders)
• Geospatial: PostgreSQL PostGIS (Haversine, ST_DWithin, polygons)
• Async processing: BullMQ for heavy tasks (imports, emails)
• Cache & rate limiting: Redis with explicit keys and user-scoped cache
• Mailing: MJML templates
• Testing: unit (Jest) + transactional E2E (per-test rollback)
• Observability: structured logs + Netdata
• Ops: Docker, Coolify (deploy), GitHub Actions (CI/CD), AWS S3 backups

Highlights

• Reusable NestJS boilerplate with strict conventions: logging, error handling, .env validation, safe caching
• PostGIS search by radius or polygon, with type filters
• CSV import pipeline: staging, validation, queueing (BullMQ), per-row errors, email report
• Cache strategy without “zombie caches” via controlled keys (a problem occuring with redis tags)
• Strong security: guards, roles, per-route & per-user cache control
• Field-friendly UX: QR codes on containers for quick scanning

Features

• CRUD for sites, areas, containers
• Geospatial search (radius, polygon, type filters)
• High-volume CSV imports with detailed reports
• Auth & roles (JWT), rate limiting, contextual logs
• Transactional emails with MJML templates
• Admin ops: automated deploys, monitoring, S3 backups

Challenges & learnings

• Load resilience: queue-driven imports avoid DB locks & keep UI responsive
• Business rule isolation: domain rules are testable without DB
• Cache safety: explicit keys & user-scoped decorators prevent bad invalidations
• Geospatial accuracy: PostGIS functions are faster & index-friendly

Tech stack

Back-end: NestJS, TypeScript, TypeORM, PostgreSQL + PostGIS, BullMQ, Redis, JWT, MJML
Front-end: React, Vite, Tailwind + DaisyUI
Quality: Jest (unit), transactional E2E tests
Ops: Docker, GitHub Actions, Coolify, Netdata, AWS S3