Now

Work

Senior Data Engineer at Ørsted in Kuala Lumpur. Building AI-integrated data pipelines for offshore wind farm operations and energy trading platforms. The work sits at the intersection of high-throughput data engineering and LLM integration — getting structured sensor and market data into a form that language models can reason over reliably.

Building

Dhi — working on v1.1 which adds multi-file edit mode (the FIM model suggests changes across two files simultaneously based on a natural-language intent). The core challenge is prompt construction: you need to give the model enough context to understand the edit target without exceeding the context window on a local 7B model.

LLM Governance Engine — adding RAGAS hallucination scoring in v1.1. Local Ollama as the evaluation judge so there's no marginal cost per inference. The faithfulness score gets stored in the audit vault 2-3 seconds after each benchmark run completes.

Writing

Working on a post about the practical challenges of running LLMs on local hardware — quantization tradeoffs, context window limitations, and when it actually makes sense to use a 7B local model vs a cloud model for a given task.

Reading

• The Pragmatic Programmer (20th Anniversary Edition) — re-reading it for the third time. Still the best single book on software craft.
• Papers on FIM (Fill in the Middle) training objectives — specifically the StarCoder and DeepSeek-Coder approaches to cross-file context.

Location

Kuala Lumpur, Malaysia. Living here since 2012.

---

Updated July 2026. Inspired by nownownow.com.