About
AI engineer building backend-first platforms, multimodal evaluation systems, and evidence-grounded decision tools.
Current work spans education AI, workforce intelligence, and public-sector delivery across Singapore and Brunei. This site focuses on inspectable systems and selected private-code case studies across LLMs, forecasting, analytics, and evaluation.
AI Engineer and ASEAN Education Program Director at Elice, leading AI feature delivery across Singapore and Brunei.
Enterprise and government stakeholder framing, multimodal evaluation, personalization, and workforce-intelligence systems.
Profile
From model behavior to production systems that hold up under delivery constraints.
I build applied AI systems with a bias toward backend clarity, evaluation discipline, and operator usability. The work I enjoy most sits where retrieval, orchestration, observability, security checks, and human review matter as much as the model itself.
At Elice, I lead AI feature delivery from scoping through technical framing and delivery coordination. Recent systems include Creator AI, a multi-service platform for course, lab, and quiz-bank generation across Discovery, Strategy, Retrieval, Generation, Validation, and HITL review; SlideBench, a multimodal benchmarking platform for AI-generated learning artifacts; and model-integration work that standardized Azure OpenAI, OpenAI Responses, Gemini, and MLAPI behind service-level controls and hardened configuration.
I also work on the operational layer: background job orchestration, admin-token protection, malware and project-code checks, regression coverage, deployment diagnostics, and Langfuse-based observability. More recent prototypes include telemetry-aware personalization for Elice's legacy LXP using Playwright probes and Bayesian Knowledge Tracing, plus a workforce-intelligence system built with Next.js, FastAPI, PostgreSQL, and Docker for governed public-source ingestion, explainable scoring, analyst dashboards, historical backfill, and forecasting.
Before Elice, at AI Singapore's AI Apprenticeship Programme, I built an LLM-based structured query and classification system for enterprise HR data and a local-first, GPU-accelerated RAG stack over more than 1,000 technical documents, with FastAPI, Pydantic, Kubernetes, and observability in the loop. Earlier consulting work at TalentKraft focused on automation and delivery, including report-generation workflows that cut analyst effort by about 70 percent.
Focus
What I tend to build.
AI platforms and orchestration
Multi-service systems spanning retrieval, generation, validation, review workflows, and reproducible artifacts.
Evaluation, observability, and hardening
Rubric judging, evidence checks, regression coverage, diagnostics, tracing, and deployment guardrails.
Backend and operator surfaces
Python, FastAPI, Pydantic, PostgreSQL, Next.js, Streamlit, internal dashboards, and explainable review workflows.
Decision-support systems
Education AI, workforce intelligence, HR analytics, finance, forecasting, and public-sector workflows.
Now
- Elice: AI Engineer and ASEAN Education Program Director since October 2025
- Cross-border product and solution work across Singapore and Brunei
- Government and enterprise stakeholder framing, including Brunei Innovation Lab Maker Phase representation
Prior impact
- AI Singapore AIAP: enterprise HR LLM workflows and local-first RAG over 1,000+ technical documents
- TalentKraft: automated more than 100 report decks and reduced analyst time by about 70 percent
- Consulting and teaching background that still shapes how systems are explained to real users
Education and research
- Georgia Tech M.S. Analytics, GPA 4.0/4.0, expected August 2026
- First-author survey in progress on hallucination in medical vision-language models with xulabs (CMU)
- AI Singapore AIAP alumnus, AWS Certified Cloud Practitioner, English and Mandarin
Portfolio scope
Case studies with a visible system behind them.
This site focuses on public case studies plus a small number of documented private-code systems across finance, document intelligence, evaluation, workforce intelligence, forecasting, pricing, and graph ML. The common thread is not the library stack. It is the translation from technical experimentation into a workflow, interface, or decision surface someone can actually use.
Each project page is intentionally concise: the business problem, outcome, key decisions, system design, and links to code, technical articles, or demos where available.
Experience footprint
The environments that shaped how I build.
- Education platforms, learning analytics, and evaluation-heavy product workflows.
- Workforce intelligence, HR data systems, and analyst-facing decision tools.
- Public-sector and enterprise solution framing where technical choices need to survive stakeholder scrutiny.
Connect
