14 years delivering software — six building interactive installations, eight extending an enterprise content platform used by banks, insurers, government and healthcare. In 2026: a four-phase AI roadmap, an AI engine and an OCR product.
An AI engine inside a legacy .NET platform, built from the roadmap above. It summarises, extracts, classifies, answers questions, bulk-uploads and reads photos, through one interface over six LLM providers, cloud or self-hosted. Clients pick a mode — Auto-Pilot, human-in-the-loop or hybrid — every call is logged with its cost, and every extracted field is checked against the source.
Template OCR with anchor matching. Confidence thresholds per field, per template and per batch decide what is filed automatically and what waits for a person. A WPF design studio, a capture station and an OCR sidecar (PP-OCR and Tesseract) share one engine. Built to replace ABBYY and Kodak Capture Pro — now adding its own verification station and template-free AI extraction, so it can be sold on its own.
Each client chooses how much the AI does alone: Auto-Pilot, human-in-the-loop or hybrid. Anything below the confidence bar, or any layout no template knows, still goes to a person.
Auto-filed or confirmed by a person, both paths are logged: who, when, which model, what was done and what was changed.
A bug isn't fixed until its root cause is proven, and work isn't done until it's checked in the live flow. Every document type is also tested in 20+ messy variations.
The AI went inside the live platform as a library — no rewrite, no new server. Database changes are additive only, shipped as idempotent, versioned scripts.
Every feature starts as a written spec, then a plan, tests first, a review, and a check in the live flow. Scope stays small on purpose.
Clients choose their AI provider and use their own API keys. Before building search across documents, I priced it: about 500 times a normal search, so it is kept for the tasks that need it.
Bachelor of Computer Science, Taylor's University (RMIT-affiliated). Silver award at Malaysia Technology Expo 2012 for FThat, an Android app. Then four months as a software engineer at Allianz Malaysia.
Client-facing project coordinator for a 10-person project team. Scoped, priced and delivered 50+ short-cycle projects — event installations, mobile apps, Facebook apps, websites — against launch dates that could not move.
Eight years extending a platform already in production: single sign-on, e-signature, self-service reporting, KPI reports, scripted data migrations. Deployment, go-live and upgrades for clients including an Islamic bank and a listed pharmaceutical group.
Wrote a four-phase AI roadmap and presented it to management. Built the AI engine and Capture Studio solo in about three months. Phases 1 and 2 done; now leading Phase 3.
