CS background. AI practitioner. Not an engineer — but close enough to the metal to matter.

I studied Computer Science at Asia Pacific University in Kuala Lumpur, specialising in Intelligent Systems. The program covered machine learning, computer vision, NLP, and systems design — which meant I spent a lot of time understanding how models work before I ever thought about deploying them for real clients.

That background led directly to my current role at Mekari Qontak, where I implement AI chatbot and Agentic AI solutions for enterprise clients across Indonesia. Most of the work happens at the intersection of product, client, and engineering: translating a business problem into a system architecture, designing prompt structures and knowledge bases, and then staying involved through deployment and iteration.

I'm not positioning myself as a software engineer — but I think the most useful people in AI product work are the ones who understand both sides. I can read a RAG pipeline and explain why it's returning garbage. I can write the prompt layer, structure the knowledge base, and talk to the engineering team about API design. That combination is what I find valuable, and what I want to keep building on.

Going forward, I'm most interested in roles where AI implementation meets product strategy — where the question isn't just “can we build this?” but “should we, and how do we make it actually work for users?”

I'm particularly interested in retrieval-augmented generation — not as a buzzword, but as the practical problem of making LLMs reliably useful over specific business data. The gap between a demo that works and a production system that doesn't hallucinate is where most of the hard work is, and it's where I've learned the most.

More broadly, I think about systems design — how components connect, where failure points are, and how to build something that can be handed off to a client and still work six months later. I find the intersection of AI product development and deployment reality more interesting than the theory-only side: what does it actually take to go from a capable model to a system a non-technical user trusts?