Singapore · Computer Science (AI/ML) · Applied AI / Software Engineering
Explore the public Mirai engineering showcase →
A recruiter-facing product and architecture case study covering the chat-first workspace, persistent business agents, Atlas developer platform, governed execution, multi-tenant security, evaluation engineering, and selected internal certification evidence.
500/500 evaluation iterations · 37 → 92/100 quality score · 49 governed tools · 201/201 active safety cases
The production monorepo remains private; the showcase exposes the engineering decisions and proof without publishing sensitive implementation or operational material.
Explore the Amethyst engineering showcase →
A recruiter-facing case study for governed company intelligence: evidence, point-in-time context, policy-enforced approvals, explicit learning writeback, lifecycle controls, and evidence-gated deployment.
CBR-PROD-00–07 complete · 200/200 health and readiness · 40-second rollback · external pilot deliberately disabled
I build systems that let software reason, act, verify, and improve across real workflows.
My current work sits at the intersection of:
- autonomous and multi-agent engineering
- production AI architecture and evaluation
- deterministic testing and replayable systems
- quantitative research infrastructure
- knowledge systems and workflow automation
- backend-heavy product engineering
I care most about software that survives contact with reality: observable, reproducible, testable, and useful to the people operating it.
|
Chat-first AI workspace and autonomous engineering system. Built around agent orchestration, worker handoffs, persistent operational context, evaluation gates, and production workflows. The goal is not another chatbot; it is an interface through which people and agents can operate software systems together. Focus: multi-agent coordination, tool use, human escalation, durable workflows, verification, agent UX |
Laptop-first autonomous quantitative research platform. A research system for falsifiable market hypotheses, deterministic backtesting, temporal-graph evidence, replayable simulations, and model-independent verification. Current research includes liquidation dynamics, forced market participants, market microstructure, and systematic setup discovery. Focus: Python, event replay, evidence ledgers, backtesting, causal research, deterministic simulation |
|
Production workflow and document-intelligence infrastructure for legal operations. Designed for high-volume intake, file processing, workflow orchestration, queueing, idempotency, auditability, concurrency, and safe human escalation across n8n, PostgreSQL, cloud services, and messaging interfaces. Focus: workflow reliability, document intelligence, deployment architecture, operations, testing under real user behaviour |
Natural language to playable 3D worlds. A browser-based generative game system that turns dream descriptions into explorable voxel environments with procedural entities, structures, physics profiles, dialogue, quests, and runtime-created game modes. Focus: generative systems, procedural worlds, agent orchestration, WebGL, game systems |
|
Governed company intelligence for AI-assisted work. Company Brain turns source evidence into approved claims, policies, metrics, entities, and point-in-time context packets. It adds two-phase learning writeback, policy-enforced approvals, immutable receipts, kill switches, recovery controls, and evidence-gated production deployment. Focus: applied AI, human-in-the-loop governance, context engineering, idempotency, audit, observability, backup and rollback |
Systems for agents to build and operate software safely. Across my projects, I design evidence ledgers, phase gates, independent review loops, reproducible builds, worker orchestration, and explicit authorization boundaries so autonomous engineering can move quickly without becoming unverifiable. Focus: agentic engineering, governance, evals, CI, release gates, reproducibility |
- Mirai Showcase — public product, architecture, and engineering case study for Mirai
- Amethyst Showcase — governed AI context, approvals, learning writeback, recovery, and production controls
- DreamCraft — generative browser-based 3D world engine
- Mototiam — motorcycle listing intelligence, scoring, and alerts for Singapore
- AMI — AWHL Market Intelligence — evidence-backed competitive intelligence platform
- PokerAI — reinforcement-learning research for No-Limit Texas Hold'em
Evidence over assumptions.
Deterministic systems over fragile demos.
Verification before expansion.
Clear contracts over hidden magic.
Automation by default, judgment by design.
I treat specifications, tests, evals, replay artifacts, security boundaries, and deployment evidence as part of the product—not paperwork added after the build.
Languages
Python · TypeScript · JavaScript · SQL · C++ · Rust
AI and research
LLM systems · agent orchestration · MCP · RAG · embeddings · evaluation systems · PyTorch · scikit-learn · reinforcement learning
Backend and data
FastAPI · NestJS · Node.js · PostgreSQL · Redis · Prisma · Supabase · REST APIs · event-driven workflows
Frontend and product
Next.js · React · Tailwind CSS · shadcn/ui · Three.js · WebGL
Infrastructure and automation
Docker · GitHub Actions · Linux · Vercel · DigitalOcean · n8n · CI/CD · contract testing · browser QA
- agentic engineering and multi-agent coordination
- LLM evaluation and reliability
- temporal and provenance-aware knowledge graphs
- quantitative market structure and forced participant behaviour
- deterministic simulation and replay systems
- integration-first knowledge products
- production architecture for autonomous software teams
Public GitHub activity is only part of the work: several production, client, and research systems are developed in private repositories.
I am interested in graduate and early-career opportunities in Applied AI, AI/ML Engineering, Software Engineering, and Forward-Deployed Engineering in Singapore or remotely.
The best fit is a team solving difficult operational problems with strong engineering discipline—not simply wrapping an API around a prompt.
Build systems that make the next system easier to build.