I like building things, trying out ideas and getting to the bottom of how they work. I'm interested in the whole process: the code, the interface, the systems underneath, and what happens when something goes wrong.
Lately, I've been building a workflow debugger in Go and Python, training retrieval models and contributing fixes to open-source projects. Have a look around.
Australia · LinkedIn
I built Checkpoint to investigate failed Python workflows without losing the inputs and responses that explain the failure. Record a run, replay supported HTTP calls, branch with a changed input, and compare the result while keeping the original run intact.
The Go engine runs the workflows, a Python SDK connects your code, and the interface brings the execution path, attempts and JSON differences together. Run history persists in SQLite; selected code and recordings can be exported as an executable reproduction. Windows and Linux releases include the interface, examples and SDK.
Try Checkpoint · Source · Architecture · Watch demo
Go · Python · React · TypeScript · SQLite
Rank Lab fine-tunes MiniLM on SciFact and compares its search results with pretrained and keyword-search baselines. It records checkpoints, resumes interrupted training and lets you inspect the documents behind each query’s score.
22.7M trainable parameters · 3 epochs · 78 updates · +3.00 pp nDCG@10 over pretrained. The explorer keeps all 53 gains and 27 regressions visible. The result is from one SciFact study; you can also train on your own documents and open the exported results in the same explorer.
Open explorer · Source · Engineering decisions · Watch demo
Python · PyTorch · React · TypeScript
An investigation workspace for engineering sources that change or disagree. Trace passages to retained versions and line numbers, compare revisions, and keep supporting and contradictory evidence attached to a conclusion. Includes a runnable fictional investigation and retrieval checks.
Project · Case study · Watch demo
React · TypeScript · Python · FastAPI · SQLite / FTS5
- Reliable Replay Lab: Python/SQLite experiments for retries, stable operation identities and recovery after a worker crash.
- LUNEA Commerce: an individual React/Express/MongoDB project with authentication, access controls and customer/admin workflows. Live demo
- Qdrant client #1563 · Merged: fixed migration retries that could lose records after a failed upload.
- REA #1829 · Merged: reused parsed module URLs during matching to avoid repeated parsing.
UTS Bachelor of Cybersecurity, expected 2027 · GPA 6.60/7.00 · Dean's List 2025 & 2026. At Zip, I built a JavaScript proof of concept and worked on Angular/TypeScript features during an internal Software Developer trial, alongside experience in regulated financial-services operations.
Working with: Go, Python/PyTorch, TypeScript, React/Angular, REST APIs, SQL, Linux and automated tests. Open to graduate and junior technology roles in Australia.
Security, infrastructure and earlier projects
- IDOR analysis and defence: a two-person investigation of object-level access control and layered defences.
- PKI and HTTPS: a private CA, X.509 certificates and TLS trust validation in a lab.
- VMware vSphere labs: ESXi, storage, networking, vMotion and High Availability.
- Database design: a two-person ERD and PostgreSQL implementation.

