About
I build AI systems designed to hold up in production: agentic pipelines, enterprise AI infrastructure, and governance scaffolding for teams deploying at scale.
I'm Vivid, based in Sydney. I build agent systems, local AI operating systems, and the governance scaffolding that lets other people run them.
Most recently that meant Breville's first MCP-based internal tool, a custom GPT sitting over more than 235,000 brand assets, which I prototyped in September 2025 and the BRG engineering team then took to production. I came to this from HCI and industrial design, where you cannot patch a product after it ships, so you learn early to ask who it fails on, and I have been asking that same question ever since even as the substrate keeps changing underneath it. Before any of that there were two startup exits in video games, one out of Shanghai and another while in Berlin, and a long stretch of living and working across twelve countries on four continents. Somewhere in the middle of all of it I did AI Product Management at Duke and an MBA at Steinbeis Hochschule Berlin, where my thesis was on AI-driven experience design.
What I work on
- Agentic AI systems: multi-agent architectures with explicit safety envelopes, LLM orchestration, evaluation harnesses
- Local AI operating systems: enterprise agentic environments with governance scaffolding, policy controls, and safety gates built into the runtime. CoworkOS (Claude Cowork) and WorkspaceOS (Codex) deployed at Breville; KimiOS (Kimi Work) for personal use, integrating a Karpathy-style second brain for longitudinal reasoning
- Privacy-first AI infrastructure: provenance-aware retrieval, longitudinal reasoning scaffolds, human-in-the-loop review gates
- Human-AI interaction: designing experiences where AI augments decision-making instead of obscuring it
Why I care about safety
Most AI demos look good on happy-path prompts and fall apart the moment you ask the hard questions: provenance, override, longitudinal change, failure modes.
That's what threads my work together: multimodal memory with chain-of-custody, multi-agent trading with execution guardrails, local agents with strict tool boundaries. The use cases differ. The discipline doesn't.
Currently building
- An AI-enabled video game, still in stealth
- Koinaku, a financial literacy app for the next generation of Indonesians, currently in early beta
- Hybrid model routing across local models (LM Studio) and OpenRouter-hosted models, including Qwen 3.x variants
- AI evaluation frameworks aimed at product decisions, not benchmark gaming
Accolades
The AI work is the project list further up this page, so what follows is the decade before it: physical products that shipped and sold in real volume, plus a few hackathon results from back when AI still mostly meant a chatbot.
Product design
- Red Dot Award 2022, Best of the Best: Breville Barista Express Impress BES876
- SCA Award 2022: Breville Barista Express Impress BES876
- Good Design Award 2022: Breville Joule Oven Air Pro
- Red Dot Award 2020: Breville Polyscience HydroPro SousVide
- Red Dot Award 2020: Breville Nespresso Creatista Pro
- IDSA 2019, Bronze: Breville Smart Oven AirFry
Software and data
- NASA SpaceApps 2017, 1st Place Australia / Top 25 Global
- NASA SpaceApps 2016, 1st Place Australia / Top 5 Global
- Chatfuel Chatbot 2017, Top 10 Global Finalist: Stake Chatbot
Education
- AI Product Management, Duke University (Pratt School of Engineering)
- MBA, Steinbeis Hochschule Berlin (Thesis: AI-Driven Application for Experience Design)
Languages
English · French · Mandarin