Available for new opportunities

I build the systems
AI products stand on.

I’m a backend engineer: I ship idempotent ingestion pipelines for messy real-world input, multi-tenant APIs where isolation is enforced at the query layer, and model integrations that fail safely instead of silently.

View projects Get in touch

Open source contributor 6 PRs merged upstream, including 3 into Reactive Resume (41k stars) · 3 packages on npm & the VS Code Marketplace

01 / SELECTED WORK

Featured projects

Multi-tenant SaaS

P-01 / minglehub

MingleHub

A multi-tenant SaaS platform for bars and venues: an NFC tap opens a live game session on every phone at the table. The venue is derived from the auth token, never the request, so cross-tenant reads are impossible by construction. Game state is server-authoritative, billing is metered and idempotent on Stripe, and 459 tests plus a CI guard keep it that way.

FastAPI Neon Stripe Clerk Supabase React

Play live

Privacy-first Data Pipeline

P-02 / financetracker

FinanceTracker

A self-hosted ingestion pipeline that turns CommBank and Westpac CSV exports into categorised spending. Every row is fingerprinted, so re-uploading a statement is a no-op. Transfers between your own accounts are matched and netted out, and a fail-closed sanitiser guarantees raw bank data never leaves the machine. 1,192 tests across the parsers, the matcher and the privacy boundary.

FastAPI SQLite OpenRouter Tailscale Vite PWA

Live demo · synthetic data

HealthTech

P-03 / nutriscan

NutriScan

Photograph a nutrition label and its macros land as validated rows in an 11-table Postgres schema. Per-user isolation is enforced by row-level security rather than application code, ES256 JWTs are verified server-side via JWKS, and a Gemini-to-fallback model chain with timeouts keeps the endpoint up when a provider degrades. Every uploaded image is sniffed, decoded under a pixel cap and re-encoded before it reaches storage or the model, and the model's answer is schema-checked before it reaches the log. Abuse is blocked at a middleware rather than merely alerted, a hijacked session on a delete spree is frozen on two layers, and trigger-fed recycle-bin rows make every deletion recoverable for 30 days.

FastAPI Neon Gemini Groq React

Live demo

Developer Tooling / MCP

P-04 / linkedin-formatter

LinkedIn Formatter

One formatting core shipped as three published artifacts: a VS Code extension, a zero-dependency CLI on npm, and an MCP server that lets AI agents apply LinkedIn's Unicode styling over stdio with exact code points. The preview card is the editor: type into the rendered post and every keystroke lands in the file already styled.

TypeScript MCP esbuild npm VS Code API

Marketplace

02 / TOOLKIT

Skills & stack

The languages, frameworks, and infrastructure I reach for.

Languages

Python
SQL
TypeScript
JavaScript

Backend & APIs

FastAPI
Node.js / Express
REST design, rate limiting, async job queues
pytest (1,650+ tests shipped)

Data & Storage

Neon Postgres (row-level security)
SQLite (WAL)
Snowflake
AWS S3

Infra & DevOps

Docker / Compose
nginx
GitHub Actions CI
Vercel / Render

Auth & Security

JWT / JWKS verification
OAuth 2.0
Cognito / Supabase / Clerk
CORS, CSP & security headers

AI Integration

Structured extraction (schema-constrained output)
Multi-provider routing (OpenRouter, Gemini, Groq)
Fallback chains & timeouts
Sentence-Transformers

03 / CONTACT

Let’s build something that holds up.

Currently open to backend and platform engineering roles and internships in Melbourne. Have an AI product that needs a real backend behind it? Drop me a line.

I build the systemsAI products stand on.