form PR-02 · project PR-0001 · Platform

ConvertX

A microservices platform for format conversion and developer utilities, built to learn how production systems are actually operated rather than just written. Two Go services sit behind a Kong Gateway Operator with MetalLB handling load balancing, and every backing service is AWS-shaped through LocalStack Pro: RDS for persistence, ElastiCache for caching, Cognito for identity with RS256 tokens verified via JWKS, SecretsManager for every credential, and CloudWatch for logs. The interesting work is the operational layer around the API rather than the conversions themselves. Both services autoscale on CPU, drain in-flight requests on SIGTERM, and are covered by purpose-built load and resilience suites that measure how the cluster behaves when pods are killed and deployments roll.

all projects source

CLEARED

PR-05.1

Architecture

as-built specification

backend: Go microservices (Gin)
database: PostgreSQL via AWS RDS
deployment: Kubernetes with HorizontalPodAutoscaler
also: Kong Gateway Operator · MetalLB · ElastiCache Redis · AWS Cognito · SecretsManager · CloudWatch Logs via fluent-bit · Prometheus + Grafana · CloudFront · LocalStack Pro

PR-05.2

Revision Log

change history

rev	note
A	Diagnosed and fixed a silent request-dropping bug on every deploy, where the HTTP server had no SIGTERM handling and severed in-flight requests
B	Diagnosed a log pipeline that reported success while shipping nothing, caused by an unmounted symlink target and a mismatched parser
C	Built load and resilience suites that defeat the gateway rate limit and the cache, which would otherwise make the results meaningless
D	Moved Grafana credentials to a generated Secret and disabled anonymous access
E	Deliberately cut planned document, image and frontend services to focus on the orchestration and reliability layer

period · June 2026 - Sep 2026

built at · Self-Driven Project

PR-05.3

Key Features

itemised

CPU-based autoscaling on both services, verified scaling 2 to 6 pods under 38,000 requests with zero failures
Zero-downtime rolling deployments: graceful SIGTERM draining plus preStop hooks, measured at 0 dropped requests through a pod kill and a rolling restart
Cognito-backed identity with RS256 tokens verified through JWKS, plus SHA-256 hashed API keys
Every credential and service endpoint fetched from SecretsManager at runtime, none in env vars or config maps
Two observability paths by design: Prometheus and Grafana for metrics, fluent-bit to CloudWatch for logs
Kong plugins for rate limiting, CORS, request size caps, correlation IDs and security headers
Four independent test layers: unit, integration (84 assertions), load, and resilience

PR-05.4

Skills

technical & applied

stack

Go
Kubernetes
Kong Gateway
Horizontal Pod Autoscaling
Prometheus
Grafana
AWS (RDS, ElastiCache, Cognito, SecretsManager, CloudWatch)
LocalStack Pro
Redis
PostgreSQL
MetalLB
Docker
Microservices

gained

Horizontal Autoscaling and Capacity Testing
Zero-Downtime Deployment and Graceful Shutdown
Observability Design (Metrics vs Logs)
Runtime Secret Management
API Gateway Configuration
Load and Resilience Testing
Service Boundary Design

demonstrated

System Design Thinking
Distributed Systems Reasoning
Debugging Silent Failures
Test Discipline
Scope Judgement
Project Ownership