I build agentic AI systems that run on real hardware

Enterprise intelligence system that monitors global news for governance risks, audits contracts and SOPs against PDPA, MAS, MOM, GDPR, and ISO 27001, tracks department scores over time, and delivers automated weekly boardroom reports — entirely on-premises. No data leaves the organisation. No cloud subscription. Zero ongoing cost after setup.

Self-hosted AI assistant with long-term memory and autonomous research. Every query is checked against stored memories using LLM-judged relevance filtering — not just vector similarity scores — so unrelated questions don't pull irrelevant context. Runs daily autonomous web research on a configurable topic and stores findings automatically. Survives machine reboots with zero manual recovery.

A multi-agent system that stress-tests startup ideas the way a room full of sceptical VCs, GTM strategists, and critics would. Nine specialised agents argue with each other across 8 domains before producing a full validation report. Launched publicly on X, Indie Hackers, and Reddit as a paid report service.

Production-quality React Native app for Singapore's 2026 MOE P6 Mathematics syllabus — built entirely from an Android phone using Termux and Expo Go. Completed phases 0–4.4: CI gate, architecture, local migration system, optimistic concurrency, change tracking, sync queue, Supabase authentication, and full sync engine. Real device bugs caught that never surface in sandboxes.

Autonomous multi-agent pipeline that generates and publishes Instagram Reels for Singapore SMEs with no human in the loop. Content ideation, scripting, production, and scheduling handled end-to-end by AI agents. Designed to scale across clients without adding operational overhead.