Six systems I shipped. AI platforms, tools that agents run on, and the tests that keep them honest.
An AI research platform where every claim traces back to its evidence. Tested blind across 46 studies. The eval harness that grades the output is the real product.
View case study →Design leadership across an auto-repair platform. AI features that moved revenue, retention, and NPS.
View case study →One support tool replacing many, for 3,200+ agents in 15 countries.
View case study →The inspection flow rebuilt for trust. 18 steps down to 12.
View case study →A Saudi Vision 2030 experience built with EY. 20K attendees from 40+ countries.
View case study →Product and research tooling for the firm that wrote Jobs-to-be-Done.
View case study →+ Other work Scarlett: a personal AI assistant that stays honest → PVARA: foundation research & design for Pakistan’s virtual-asset regulator → MobiMatter: eSIM checkout conversion, +10% in testing →
As products learn to think, design has to become auditable.
I design AI-native products, and I publish the research behind them. Every number on this page links to where it came from, so you can check it instead of trusting me. I built Articos from nothing as its lead researcher and designer, and shipped product work at AutoLeap, Careem, and EY.
One protocol runs everything here, from AI-native product design to published research. Every step leaves a record.
Ship evidence, not opinions. Make the work auditable, and it becomes impossible to argue with.
■ Operating principle · how I run product research
I research how people and AI actually behave together, then design around what I find.
Each paper begins with a question.
Each question uncovers a system.
Each system shows new possibilities,
and new responsibilities.
A first-principles architecture for LLM-based synthetic UX research
The architecture behind Articos. AI stands in for research participants, but every finding traces to evidence and carries a confidence score. Spotting themes is easy. Being right about them is the hard part, and that is what this buys: 42% precision against 6.7%.
A CFG-based translation system from video and audio to PSL gestures
A multi-stage pipeline that turns English and Urdu media into validated Pakistan Sign Language gestures. Built to lower communication barriers for Pakistan’s deaf community.
2024 IEEE Xplore ↗ ⟶
Beyond this point, there are no more pages.
Only signals, intentions, and systems
waiting to be understood.
Have a difficult system, strange idea, or research problem worth investigating?
Start a conversation ↗■ Available for select projects
Remote, worldwide