Logistics SaaS Platform
A multi-service logistics platform with centralized identity, role-based access and real-time data sync, serving about 1,000 daily users.
- Java
- Spring Boot
- Kafka
- Debezium
Harry Le · Lê Ngọc Đàn
Backend & Platform Engineer (Java) and Team Lead · Spring Boot, Kafka, Keycloak, React
I build the platform behind a logistics SaaS used by about 1,000 people every day: identity, microservices, micro-frontends and event-driven data.
What I work on
Six years on one production platform- #01
Led the migration of a logistics/freight SaaS platform (~1,000 daily users) from a monolith to microservices: system design, database migration, and integration with Keycloak, Kafka and internal services.
- #02
Designed and maintain the shared platform layer every service relies on: identity and SSO (Keycloak, OAuth2/OIDC), role/permission/data-scope access control, data access, file storage, webhooks and service federation.
- #03
Built the web frontend as micro-frontends with Webpack Module Federation: one host shell loading independently deployed React apps per domain (CRM, freight, HR, transport, documents, e-customs).
- #04
Implemented event-driven data flows with Apache Kafka and Debezium CDC on PostgreSQL, JPA/Hibernate and Liquibase.
- #05
Took the CRM product from design to production use; now run and maintain it, and develop the platform's mobile app.
Selected projects
All projects →ai-ocr
HTTP API that turns Vietnamese business documents (VAT invoices, receipts, customs declarations, delivery orders, CVs, spreadsheets) into structured JSON, with the heavy OCR/LLM core on Modal GPUs.
- Python
- Tornado
- uv
- Tesseract
vn-trading-agents
Multi-agent LLM tool that analyzes Vietnamese stocks, from analyst reports to a bull/bear debate and a risk-checked trade decision.
- Python
- Anthropic Claude
- vnstock
- Pytest
Writing
All posts →The guards, gates, loop detectors and state files I put around a coding agent so long background jobs stay safe, and what I could and could not verify.
When CDC is worth it, the outbox pattern, and the traps I hit with slots, snapshots, ordering and duplicate events on PostgreSQL and SQL Server.
How I wrapped IMAP in a small TypeScript MCP server: the tool design, where the safety line sits (drafts, never send), and what I would change.
Have a backend or platform role in mind?
Send me the job description, the company and the salary range. If it fits, I will reply within a few days.