Built, shipped, owned.

Problem

Each site needed to run autonomously — collecting equipment telemetry, alerts, and operational events into its own database — while feeding a central server for cross-site analytics, executive dashboards, and disaster-recovery backup.

Approach

• ▹Java + Angular application deployed as independent site instances, each with a local PostgreSQL/TimescaleDB store for time-series equipment data.
• ▹SymmetricDS hub-and-spoke replication: sites act as leaf nodes, replicating writes to a master server asynchronously with conflict resolution and staged onboarding for new sites.
• ▹Ingestion pipeline for device-generated JSON payloads — alerts, telemetry samples, event history — with unit conversions, custom alert-threshold evaluation, and email notifications.
• ▹Grafana dashboards over the central database for cross-site fleet analytics; periodic sync from a secondary MS SQL Server for machine metadata.
• ▹RAG conversational agent on top of live metrics + a PDF manuals knowledge base (embeddings) so on-site staff can ask natural-language questions over equipment history.

Result

{TODO: e.g. 'X sites onboarded · Y million events/month replicated · zero-downtime central failover'} — fill in.

Problem

Replace a fragmented stack (separate course, finance, and certificate tools) with one Angular + Java platform that owned the full learner journey — from enrolment and payment through to government-issued credentialing.

Approach

• ▹Course authoring + delivery surface: lessons, assessments, learner progress, certificate generation on completion.
• ▹Mailouts / lifecycle email: enrolment confirmation, due-date reminders, expiry warnings, certificate dispatch.
• ▹Finance module — invoice generation, reconciliation, refunds — with two-way Xero accounting integration so finance staff never re-key a transaction.
• ▹Eway payment gateway integration for card-on-file enrolments and recurring training subscriptions.
• ▹USI (Unique Student Identifier) integration with the Australian government registry for compliant credential issuance — a hard regulatory requirement for any RTO.

Result

{TODO: e.g. 'X learners onboarded · $Y/month routed through Xero · USI verified at enrolment in <Z seconds'} — fill in.

Problem

Make LLM features cheap, accurate, and testable enough to ship to production: not a demo, not a side-tool — features that real users hit.

Approach

• ▹RAG retrieval over operational corpora — equipment manuals, internal documentation, historical tickets — chunked, embedded, and served behind product-specific assistants.
• ▹OCR pipeline for ingesting paper-trail documents (forms, certificates, invoice PDFs) into structured, queryable records.
• ▹MCP-style tool layer giving the LLM controlled access to live product APIs — read metrics, draft updates, surface evidence — instead of free-form generation.
• ▹DPO (Direct Preference Optimisation) fine-tuning on collected agent responses so the assistant matches the company's voice and business rules without prompt-bloat.
• ▹AI-assisted test automation — generated UI/API tests, regression checks, and assertion synthesis layered onto the existing release pipeline.

Result

{TODO: e.g. 'cut analyst look-up time from X→Y · regression suite expanded by Z% with no headcount add'} — fill in.

Problem

Standard form builders couldn't express the spatial, drawing-on-paper feel finance staff wanted. Needed a real canvas surface — drag, snap, layer — without dropping out of Angular.

Approach

• ▹Paper.js canvas embedded in an Angular 19 component, with a state bridge so the canvas tree round-trips cleanly to the existing form-definition model.
• ▹Custom snapping, alignment guides, and multi-select; serialisable canvas state so designs version and review like the rest of the product.

Result

Niche internal tool — small user base by design. Included here as a clean example of bringing a non-React / non-DOM rendering library into a modern Angular app without leaks.