Skip to content

Hello, I’m Charles, a software engineer based in Melbourne. I run Backbone Engineering, where we help introduce growing businesses to the complex world of data systems.

Full Stack Software Engineer

Available from November 2026 · hybrid || remote

Also as WordMarkdown

Timeline

Years 2024 – 2026

Showing 3 of 5

Software Engineer (Contract)

Backbone Engineering · Feb 2026 – present

  • Software
  • Self-employment

Software contracting firm, engaged full time from April 2026 to November 2026 by a scaling Australian importer–distributor in the flooring industry (Synteko Australia).

  • Designed, developed and delivered a B2B/B2C commerce platform from a Shopify proof of concept to a custom Medusa.js build. The store integrated with the client's existing Jiwa ERP, was delivered progressively via CI/CD and regular showcases, and deployed on DigitalOcean App Platform
  • Modelled the client's operational database and an analytics data warehouse for operations, revenue operations and reporting.
  • Built internal TypeScript and Node.js tools that generated suggested stock orders from historical sales trends, live pricing, stock levels and projected demand. Deployed as part of a Next.js internal platform application to inform stock policy and assortment
  • Designed and implemented an internal pricing calculator to calculate how changes in pricing would affect margins across the range given historical data. This also allowed saving the prices of competitors to see opportunities to price more competitively.
  • Modelled, deployed and migrated a postgreSQL database from a mySQL backup, reducing the time cost of string search at point of sale from an average of 11500ms to 60ms using a server-side fuzzy search. This resulted in >200 minutes saved weekly in time to serve customers by Synteko's warehouse and retail staff.
  • Worked directly with the managing director and senior staff from discovery and requirements through UI/UX design, delivery and support, and trained warehouse, sales and administrative teams on the new platform.
  • Scoped and reviewed work for two part-time engineers brought on through the practice.

TypeScript · Node.js · Medusa.js · Next.js · PostgreSQL · AWS · DigitalOcean

Read the full CV

Selected work

  • Trade and retail storefront developed for a client

    Scheduled to be deployed November

    Architected and developed a B2B/B2C storefront with trade and reseller accounts and pricing, live stock levels, click & collect, payment integrations, and returns.

    Medusa.js · TypeScript · Vercel · AWS · Jiwa

  • TimberWatch

    Timber-health monitoring system, developed with electronics engineer Victor Liu. A microcontroller within a TimberWatch unit reports via UDP to a central web server on DigitalOcean, which auths and writes to PostgreSQL; the data feeds visualisation, with model training planned once a sufficient dataset is gathered. I developed the Rust that powers authorisation and the web server.

    Rust · Axum · PostgreSQL · UDP · DigitalOcean

  • charlesaquilina.com

    This site. Strict TypeScript behind a CI gate of lint, type, test and accessibility checks; the CV is one typed record rendered to web, PDF, Word and Markdown. The site has support for blogs, and first-party analytics with no cookies.

    Next.js · TypeScript · PostgreSQL · Vitest · GitHub Actions

  • Elastic social media processor

    University team project (4 people). Harvested Bluesky, Mastodon and Reddit posts on the Melbourne Research Cloud for Australian federal election analysis: serverless ingestion and ETL into Elasticsearch, HuggingFace sentiment models and a REST API. Owned the cluster, deployment, schema and CI/CD.

    Python · Kubernetes · Fission · Docker · NLP · HuggingFace · Elasticsearch

  • HPC processing of large social media datasets

    Generating aggregation metadata and trends in >40 Terabytes of enriched tweets + Mastodon posts. Processed terabytes of NDJSON with MPI on the University of Melbourne's Spartan cluster: ~80% parallel efficiency and the cohort's lowest wall-clock time on datasets over 8 GB, via byte-offset preprocessing.

    Python · mpi4py · SLURM

Latest writing

All posts

How this site is built

  • Every push runs lint, formatting, type checks, tests and a production build in CI.
  • TypeScript with every strict compiler flag on.
  • An automated accessibility check on every page.
  • The CV is one typed record, rendered to this site, PDF, Word and Markdown.
  • First-party analytics: no cookies, no identifiers, no stored IP addresses.
  • Server functions run in Sydney, beside the Postgres database.