Backend Engineer — Java/Spring Boot & Node.js · Bangkok, Thailand

I build backend services for production traffic. Most recently a real-time analytics platform sustaining 10M+ requests/day at sub-100ms response times, built on Kafka and Redis. Two platforms I'm building at One Inspector and Consultant, my own consultancy: Sentinel IoT Platform, an industrial monitoring system in Java 21 / Spring Boot with MQTT ingestion, circuit-breaker resilience and GitOps deployment to EKS; and Construction OS, an AI-native construction management platform currently in development.

Languages

Backend

Data & Messaging

Infrastructure & CI

Observability

Frontend & Testing

The systems below were built under commercial engagements, so the source is proprietary and not published here. Happy to walk through the architecture and trade-offs in conversation.

Real-time analytics platform — Kafka, Redis, Docker Sustains 10M+ requests/day. Redis caching layer brought API responses under 100ms; Kafka-based ingestion decoupled the write path from reporting reads so traffic spikes stopped degrading dashboard performance. Collapsed client reporting from a multi-hour batch job into on-demand results.

Distributed chat service — WebSocket, Pub/Sub Pub/sub fan-out keeps message delivery consistent across nodes instead of pinning users to a single server.

High-throughput notification pipeline — Kafka Event-driven delivery with retry/backoff, dead-letter handling, and idempotent sends.

AI-native construction management platform built for enterprise scale. Currently at Stage 1 (BUILD) — architecture and specs are settled, implementation is underway.

• Modular monolith over microservices — one NestJS deployable holding all domain modules (identity, tenant, project, BoQ, procurement, site-ops, finance, equipment, workforce), with separate services only where a language or throughput boundary justifies it: Python/FastAPI for LLM gateway, embedding, OCR and transcription; Go for analytics, knowledge-graph and MQTT IoT ingestion
• Multi-tenancy — shared database with tenant_id and PostgreSQL row-level security, chosen over schema-per-tenant and documented as an ADR
• Offline-first clients — Serwist PWA with IndexedDB on web, Drizzle over expo-sqlite with a sync queue on React Native, reconciling through Kafka; built for sites with no signal
• Monetary correctness — DECIMAL(19,4) in PostgreSQL, decimal.js and Python decimal in application code, floats banned by lint rule
• Event bus — Kafka with Avro and Schema Registry under BACKWARD_TRANSITIVE compatibility; Temporal for long-running workflows
• Auth split by user reality — phone + SMS OTP via AWS SNS for site workers, Keycloak OIDC with RS256 JWT for office staff
• Polyglot persistence — PostgreSQL + TimescaleDB, ClickHouse (analytics), Neo4j (knowledge graph), pgvector + OpenSearch (retrieval), MinIO (objects), Redis
• Operational discipline — PgBouncer in transaction mode with direct PostgreSQL connections prohibited, OpenTelemetry → Grafana/Loki/Prometheus, secrets via Vault and AWS Secrets Manager, ~60 ADRs and a 100% line/branch coverage mandate

Industrial IoT monitoring platform — MQTT sensor ingestion, threshold alerting, multi-channel notifications, live WebSocket dashboard, and a full observability stack.

Measured: cache read path sustains 1,000 req/s (60k+ ops/min) at p95 112 ms / p99 187 ms under k6 load test, against SLO targets of p95 < 200 ms and p99 < 500 ms.

• Ingestion — Mosquitto MQTT via Spring Integration, with DLQ routing for invalid payloads and unknown devices; Kafka event streaming with Avro schemas and Confluent Schema Registry
• Resilience — Resilience4j circuit breaker and retry; when Postgres is unavailable the dashboard stays live off Redis while writes queue to a Redis replay list, drained every 30s once the breaker recovers
• Data — PostgreSQL 16 with monthly range partitioning, hourly aggregates, row-level security for multi-tenancy, Flyway migrations; CloudNativePG + Barman for HA replication and WAL archiving
• Realtime — Spring WebSocket with Redis pub/sub fan-out across nodes
• Observability — Prometheus, Grafana, OpenTelemetry/Micrometer, Jaeger tracing, structured JSON logging with MDC request correlation, multi-window burn-rate SLO alerting
• Testing — JUnit, Mockito, Testcontainers, Pact contract tests, JMH benchmarks, Pitest mutation testing, ArchUnit, schemathesis fuzzing, k6 load tests
• Deploy — Docker Compose locally; EKS via Terraform and Helm, GitOps with ArgoCD, blue/green and canary rollouts with Argo Rollouts, event-driven autoscaling with KEDA, Velero backup/DR
• Frontend — Next.js 14 App Router, Tailwind, Recharts, react-query, Zustand, list virtualization

RESTful service integrated with the LINE Messaging API, handling webhook events and message delivery.

Full-stack platform for user and content management. Currently migrating the frontend from server-rendered EJS to React + Tailwind CSS.

E2E suite for a cinema booking platform, covering language switching → cinema selection → movie selection → showtime verification → seat selection.

• Page Object Model — selectors and interactions encapsulated per page class
• Fixture-driven test data, decoupled from test logic via JSON
• Mochawesome HTML reports, generated in CI on every push and pull request
• Retry strategy for flaky network-dependent tests

Secure remote voting system with live result updates, backed by a microcontroller client. Firebase Authentication, Cloud Functions, and Realtime Database.

IoT automation that reads environmental sensor data and drives microcontroller-based window control logic.

Google Cybersecurity · Prompt Engineering & Agentic AI (AIS Academy) · Secure Coding: OWASP Top 10 · Secure Coding on Frontend