Final Year Data Science & Analytics student at NUS (graduating June 2027), building explainable AI systems, LLM-powered agents, and production-grade ML applications.
Currently interning at HTX (Home Team Science and Technology), exploring vector databases and AI system design in a security-focused environment.
I focus on:
- Agentic AI workflows and multi-agent orchestration
- Retrieval-Augmented Generation (RAG) systems
- Hybrid ML + LLM architectures
- Explainable company intelligence systems
- 🌱 Pursuing a B.Sc. (Hons) in Data Science & Analytics at NUS
- 💡 Bridging AI/ML engineering with product thinking — I like challenging engineering decisions, not just executing them
- 🚀 4 internships across fintech, fraud detection, and AI entrepreneurship: CPF Board, Superbank, Crayon Data, HTX
- 🏆 1st Place, NUS Datathon 2026 (IntelliCompany AI)
- 🏆 Finalist, NUS SDS Hackathon 2025 (Top 3/40)
- 🏆 Finalist, NUS SDS Datathon 2025 (Top 5/75)
- Careerlingo — Duolingo-style AI career coach (LinkedIn Career Trailblazer Hackathon), building the Scout agent
- Vector database evaluation with Milvus
- AI system design for security/Trust & Safety contexts (HTX internship)
- Bernard Builds — freelance web design studio, increasingly positioned around AI-powered websites
-
Data Science Intern @ HTX (May – Aug 2026) → Working on AI system design and vector database infrastructure in a security/Trust & Safety context.
-
AI Entrepreneur Intern @ Crayon Data (Dec 2025 – Jan 2026) → Built LLM-powered agentic workflows transforming unstructured merchant offers into schema-compliant datasets.
-
Operations (Data Science) Intern @ Superbank (May – Aug 2025) → Engineered 200+ fraud detection features in Python/SQL, optimised queries on 50M+ records (~40% faster), built real-time monitoring pipelines in Snowflake.
-
Publicity Head @ Alumni Relations Committee, NUS Science Club (2025–Present) → Leading publicity and design for alumni-facing events.
Explainable AI-powered company intelligence platform combining classical ML clustering with grounded LLM reasoning.
Architecture:
- K-Prototypes clustering on 8,559 companies, 20+ engineered features
- Structured retrieval (15 records/query) feeding a guardrail-constrained Llama 3.3 70B (Groq, temp 0.1)
- Strict dataset-only grounding — zero hallucination policy
- Full-stack: React + FastAPI, Dockerised
Why it matters: demonstrates hybrid ML+LLM design, explainability-first engineering, and end-to-end shipping.
- Grounding over hallucination
- Classical ML + LLM hybrid architectures
- Explainability before automation
- Low-temperature deterministic outputs for business contexts
- Clear guardrails and retrieval constraints
- Containerised, deployment-ready systems
Safe · Interpretable · Scalable · Cost-aware
