For Niki Forecast · Abtrace · 20 August 2026
A few things we didn’t get to
Thank you for the call today. A short follow-up on the projects we didn’t get to during our conversation.
01
Dododo
The story I mentioned on the call. After the pilot, we are rebuilding Dododo around a conversational assistant rather than a static plan. I designed it in code with Claude, connected Supabase so the prototype runs on real data, and tested how the model behaves across different child profiles: a three-year-old with sensory needs gets a very different experience from a seven-year-old with ADHD.
02
OED AI
Model limitations, handled by design.
The AI quotation finder takes several minutes to return an answer. The platform convention was to use a modal, but with that kind of latency, a modal traps the user. So I embedded the assistant in the page: it sits quietly until needed, the query keeps running, and the researcher can carry on working. The model’s speed was the limitation; the design absorbed it.
03
STIX
A pre-AI medical product I would build differently today.
An app that takes a short video of a urine test strip, giving the model multiple frames to analyse for a more accurate reading, so clinicians can get results faster and more consistently.
I ran a six-hour stakeholder workshop with the team and clinicians: three personas, journey maps, prioritisation, and then a clickable prototype handed over for clinical validation and development.
We built it before the current generation of AI tools. Today, I would prototype it against the real model from day one, and the road to an MVP could be a week rather than months.
It was great speaking today, I’m looking forward to continuing the conversation and hopefully meeting again in the next round.