Java · Middleware · Kafka · Cloud
Software engineer — backend Java, middleware and real-time data
I build backend services and integration layers for telecom systems. Four years of Java: at Vocus, the middleware carrying orders between Australia’s national broadband network and the providers reselling it, and before that the Kafka pipelines and microservices at Globe Telecom that carried subscriber data for more than 70 million people.
Currently
Vocus (via Tech Mahindra) · Jul 2026 – Present
Middleware and web-service integration for Vocus's wholesale broadband business — the systems carrying orders between NBN, Australia's government-owned national network, and the retail providers reselling it — across the full lifecycle from interpreting a requirement through development and testing to production support.
Full backgroundBackend
Streaming, middleware and API security — the same problems as the day job, taken apart in repositories you can build and run. Each one is covered by a suite that needs no broker, no backend and no Docker.
A pipeline that parses raw usage events, de-duplicates redeliveries, aggregates volume per subscriber into event-time windows, raises threshold alerts, and dead-letters anything unusable. The whole topology is tested without a broker.
Read case study
23 tests
Middleware · Camel · Spring Boot
One REST interface over a SOAP backend and a REST one
A middleware layer that gives callers one interface over two very different provisioning backends, and answers the questions a gateway exists for: what to retry, what to give up on, and what a caller gets back when the far end is down.
Read case study
26 tests
API security · Spring Security
Personal data behind authorization that actually checks the record
A subscriber API carrying personal data, with JWT authentication, object-level authorization, field-level encryption, per-caller rate limiting and an audit trail that records refusals as well as successes.
Read case study
38 tests
Products
A resort booking engine, a coffee shop's reservations and menu, and a store's inventory and till. Each one is the half of the product worth building carefully — the rules that decide what can be sold, and what happens when the answer is no.
The booking half of a resort site: per-night availability across room types, seasonal and length-of-stay pricing, and refusals that name the night that blocked the stay rather than shrugging.
Open demo
33 tests
Table reservations with turn times and table combining, plus a menu whose modifiers actually affect the price. Slots are offered only when a real table can seat the party for as long as it will be sitting there.
Open demo
31 tests
Inventory, costing and billing for a neighbourhood shop. Quantities are derived from stock movements rather than stored, costs are weighted-average, and receipts extract VAT from prices that already include it.
Open demo
33 tests
One CRM across the resort, the coffee shop and the store. Scored identity resolution rather than a customer table, segments derived from transactions rather than tags, and consent that has to be proved before anybody can be contacted.
Open demo
55 tests
Lab
Pitch detection, sample-accurate scheduling and chord-shape search, written from the algorithms up with no audio libraries. Same instinct as the backend work, somewhere you can hear the result.
src/lib/audio/yin.ts on a synthetic A3. The dip is the period; the dashed line is the detection threshold.A chromatic instrument tuner that listens through the microphone and reports the note and its deviation in cents. The pitch detection is a from-scratch implementation of the YIN algorithm — no library — and it is accurate to under a cent on synthetic signals.
Open demo
A step sequencer whose timing does not drift, does not stutter under load, and does not fall apart in a background tab. Every drum sound is synthesised from oscillators and shaped noise, so the whole instrument ships as zero bytes of audio assets.
Open demo
An interactive fretboard that finds playable fingerings for any chord in any tuning by searching the neck under physical constraints. It ships no chord dictionary, which is why it works in tunings nobody has tabulated.
Open demo
How it is built
Across four repositories, Java and TypeScript
No broker, no database, no Docker, no backend
Across the guitar range, on synthetic tones
Proven over 200 irregular polling intervals
The same rule runs through all of it: keep the logic that matters away from the infrastructure it happens to run on. A Kafka topology is a pure function from config to a topology object, so the whole pipeline runs under a test driver with event time controlled to the millisecond. Pitch detection takes an array of samples and a sample rate. Neither knows what it is deployed into, which is why the awkward cases — a duplicate delivery, an event arriving after its window closed, a backend that fails twice and then recovers — are ordinary unit tests instead of a QA exercise.
Contact
The quickest way to reach me is email. I read everything.