Full-stack engineer at NCS Group, deployed inside government agencies for six years: tax refunds at IRAS, work passes at MOM, cross-agency APIs at DOS. I take systems from requirements with business users to production rollout. Outside work I build governed LLM systems on AWS: multi-model routing, eval gates, audit trails, all live below.
Ship production-grade apps, fast.
Open-source web and mobile frameworks built so good ideas stop dying in plumbing. Describe an idea to an AI coding agent, point it at one of the frameworks, and it ships a real, live AWS prototype with the infrastructure, auth, database, CI/CD, and security scanning already wired. The frameworks require the agent to turn the brief into a spec and write tests alongside the code, and nothing deploys until every requirement is covered by a passing test.
LiveA multi-tenant AI tax assistant platform for tax officers: each department gets its own workspace with a document-grounded assistant (RAG with clickable inline citations) and a visible step trace. One observed gateway routes every call across six OpenAI and Anthropic models by deterministic rules, logging latency, tokens, and cost, with cross-provider fallback. Officers compose their own lookup, template, and calculator tools (code runs in a QuickJS WASM sandbox), prompt versions activate only past an LLM-judge eval gate, and governance-as-code enforces PII redaction, a per-call cost ceiling, and a full audit trail.
LiveAI driving companion for Singapore: an Expo app on a serverless NestJS API (Lambda + Neon Postgres). Surfaces live ERP, traffic, weather, and parking, tracks drives with background GPS and in-drive voice alerts for gantries and incidents ahead, itemizes fuel and ERP costs after every trip, and takes voice or text questions through a Claude assistant grounded in your vehicles and driving history.
LiveAI insurance policy checker for Singapore. Drop in a policy PDF (read in your browser, never stored) and get a plain-language summary, a coverage breakdown, and a curated fine-print checklist. A LangGraph draft-verify-finalize loop grounds the model: deterministic code checks every quote against the document and re-drafts what fails, so nothing unquotable is shown.
Six years embedded in agency delivery teams: requirements with business users, build, test, production rollout. The LLM systems above run the same loop: spec, tests, eval gates, then deploy.
NCS Group, Singapore
Refund module of IRIN3, IRAS's cloud-based tax administration platform. Embedded with IRAS delivery teams.
NCS Group, Singapore
Central API Management platform, the cross-agency API layer between MTI and the Department of Statistics.
NCS Group, Singapore
IWPS and WPOL at the Ministry of Manpower. Singapore's platform managing the full work pass lifecycle.
NCS Group, Singapore
Service request enhancements on Singapore's work pass platform.
Most engineers pick a side: Java or .NET. I ship in both, plus the governed LLM stack on AWS.
Multi-model routing · eval gates · PII redaction · audit trails · AWS
.NET Core · Spring Boot · Angular · TypeScript
Apache Kafka pipelines from ingest to recovery
Fortify · SonarQube · Dependency Check inside CI/CD
Oracle SQL · MSSQL · Entity Framework Core · Hibernate
Cross-agency REST platforms on WebLogic with JNDI security
Software Engineering, National University of Singapore
I'm open to roles deploying AI systems into production, especially in regulated environments. Six years inside government agencies on .NET Core, Java, and TypeScript; now shipping governed LLM platforms on AWS. Based in Singapore, open to travel.