AI stack
- Runtimes
- Claude Code the main driver, on Opus
- Codex GPT-5.5 and GPT-6, plus DeepSeek Flash through a custom provider
- opencode open models: GLM-5 Turbo, DeepSeek, local Gemma and Qwen on Ollama
- Terminal
- Built by me
- MCP
- context7 current library docs
- Chrome DevTools drive and inspect a real browser
- PostHog who actually used it
- Neon Postgres branches
- Vercel deploys and logs
- Lazyweb real app screens before designing
- Skills, mine
- workbenchdevourscourtotalityisomorphpotentialgauntletbedrockresearch-reportmemory-scriberskill-distiller
- hatsautoreviewbirdwatchoverseer
- extratrail-scribertestament
- Skills, adopted
- impeccabledesign-taste-frontendemil-design-engmotion-designfrontend-designlazywebgsaphumanizerimagegen
- The loop
- 01vision keep what the project is for written down
- 02devour / scour understand what is actually there
- 03potential / decide choose what happens next
- 04build Claude Code and Codex, often side by side in herdr
- 05bedrock / gauntlet / autoreview prove it holds
- 06memory-scriber carry it into the next session
Projects
Deep dive →Open source
Experience
- 2026 / NowAI Engineer
Atos
Building a security assessment and evidence platform for authorised engagements, taken from research prototype to a packaged Windows pilot: multi-lane assessment runs, findings triage, release gating, reporting, remediation tracking, and verified retest. Shaped the assurance method underneath it, evaluating AI systems across behaviour, prompts, models, tools, permissions, and real-world actions, with known-answer testing against fabricated targets.
- 2026Frontend / AX & Agentic Engineer Intern
Fantopy AI · Founding Team
Designed and built the frontend of a consumer fantasy-football platform used through the 2026 World Cup: a 115k-line React 19 app with 78 screen files, PWA and Capacitor builds, and a motion system of 22 GSAP choreographies. Took the World Cup product from non-functional to live in one overnight push (13 PRs, 8 root-caused launch bugs). Built Fantopy Orchestrator, an issue-to-draft-PR agent factory with human-only merge authority: 65 issues into 38 PRs, 25 merged, in 16.6 days.
The full record→ - 2025 / 2026Automation Developer
NCS · DBS Core Banking (via RiDiK)
API and integration test automation for DBS core banking (account opening, CASA, deposits) using Karate DSL and Postman. Validated business rules, transaction flows, and backend service responses against release procedures.
- 2024 / 2025Software Test Engineer Trainee
Visa · Digital Marketing (Singapore)
Pioneered an agentic-first Playwright framework with Claude Code, MCPs, and context chips. Authored the team's agentic-first development guide. Designed specialist agents (Calibrator, Debugger, Test Creator). Cut execution time up to 80% with GenAI assertions.
- 2022 / 2024Staff Assistant
Singapore Police Force
Automated end-to-end generation of 450 personalized career transition reports using ChatGPT, Python, and UIPath. Managed Financial Planning and Courses initiatives. Revitalized microsites, newsletters, confirmation emails. PowerBI and Plotly reporting.
- 2021Video Editor Intern
Easy Video
Cut footage into broadcast-ready pieces: clips, graphics, dialogue, sound effects, and music edited into a coherent, visually appealing story. Worked closely with the production team on requirements and creative direction. Assisted high-key live stream events including the One Public Service Observance Ceremony (OPSOC) and the SPF National Day Observance Ceremony.
Education & certifications
- Professional Diploma in Software DevelopmentNational University of Singapore, ISS2024 / 2025
- Diploma in Digital Visual EffectsNanyang Polytechnic2018 / 2022
- AI200: Applied Machine LearningHeicoders Academy2024
- AI100: Python & Data VisualisationHeicoders Academy2023
Skills
- AI-Native
- Claude CodeCodex CLIMCPsAI SkillsAgentsSubagentsSpecialist AgentsContext EngineeringPrompt EngineeringAgent OrchestrationAgentic Workflow ArchitectureHuman-in-the-loop OrchestrationMulti-agent ReviewVerifier TaxonomyModel RoutingMemory EngineeringEvalsGenerative AI
- Security & Assurance
- AI Security AssessmentAI AssuranceKnown-answer TestingFindings TriageRelease GatingRemediation & RetestHuman Approval Gates
- Languages
- TypeScriptPythonSwiftJavaRustJavaScriptHTMLCSS
- Frontend
- ReactNext.jsAstroSwiftUIElectronTauriSvelteTailwindFramer MotionGSAPMDXAngularVitePWACapacitorMapbox GL JSMotion Design
- Backend & Realtime
- Node.jsFastAPISpring BootBetter AuthGoogle Apps ScriptSSETelegram BotsTelegram Mini AppsGmail APIPrivySolanaSentryLangchain
- Data
- PostgresNeonDrizzleMongoDBMySQLRedisSQLiteOCRImage Hashing
- Testing & QA
- PlaywrightSeleniumKarate DSLPostmanVitestpromptfooUIPathTest AutomationIntegration Testing
- Tooling & Infra
- GitGitHub ActionsGitHub AutomationWorktree IsolationBranch / PR HygieneDockerDocker ComposeLinux Server AdminTailscaleWindows PackagingVercelHomebrewmacOS DistributionGhosttyTmuxQuartoVSCode ExtensionsJiraZephyr
- Product & Brand
- Brand SystemsVisual DirectionProduct LaunchStoryboardingGame DesignProduction HardeningAccessibility
- Misc
- After EffectsCinema 4DPhotoshopIllustratorPremiere ProCapCutCanvaPowerPointExcelWord
- Approach
- Systems-meets-Imagination0-to-1 Product BuildingAmbiguity NavigationRapid PrototypingTaste-level IterationTechnical EvangelismSelf-directed LearningSystemic ThinkingTechnical WritingSolo ShippingAgileScrum