My own LLM models
Training and experimenting with language models from scratch to really understand what happens under the hood.
Hello there
Software and cloud solutions architect. I've spent 25+ years building things with .NET and the cloud — these days I spend most of my time teaching machines to be useful. Off the keyboard you'll find me behind a camera or reading about markets.
About
Riza Marhaban is a software and cloud solutions architect with over 25 years of experience in IT infrastructure and software development. He is a .NET developer and a former Technical Evangelist at Microsoft. His professional interests focus on disruptive technologies, particularly within the IT ecosystem, AI, green technologies, global economy, and geopolitics.
Beyond his technical work, Riza has a strong interest in photography, cinematography, music, and investing. He is currently working at the Centre of Excellence in Modelling and Simulation for Next Generation Ports at the National University of Singapore, where he supports researchers in developing simulation software, digital twin software and AI related experiments.
Lately outside of his office work, much of the focus has shifted towards applied AI engineering: building his own AI LLM models from the ground up, fine-tuning open models with Unsloth, and developing agentic AI systems with the Microsoft Agent Framework. He also builds cloud-native distributed applications with Aspire (foremerly known as .NET Aspire), and works in the spirit of a Forward Deployed Engineer (FDE) — sitting alongside the people who will actually use the software, and shipping solutions that fit their real problems.
Currently
Training and experimenting with language models from scratch to really understand what happens under the hood.
Adapting open models to specific domains, fast and cheaply, without needing a data centre to do it.
Designing multi-agent systems and tool-using assistants with the Microsoft Agent Framework.
Composing cloud-native, observable .NET services that are genuinely pleasant to run and debug.
Working shoulder to shoulder with users and researchers, turning messy real-world problems into shipped software.
Say hello