01 / AI TEAMWORKPick up a handoff without starting the search over.
When work passes between product, engineering, design, and operations, information and progress can end up scattered. We are developing more than 100 AI assistant roles with different specialties, connected to a range of AI models and services.
We are also shaping how work is assigned, projects kept separate, priorities set, and results checked by another assistant. This way of working is still being developed; it does not mean every task can be automated.
100+ assistant specialtiesMultiple AI modelsResume after a pauseA second assistant reviews the result
Progress: defining assistant roles and handoffs across different disciplines.
02 / PROJECT P · IOSCarry walk memories into a 3D home for your pet.
Project P puts your pet at the heart of the everyday story. Save moments from a walk, then keep building their 3D home through the time you spend together.
The current focus is connecting pet profiles, walk records, stories, and a 3D home on iPhone.
Pet companionshipWalk stories3D pet home
Progress: refining the iOS experience and prototype.
Discover Project P ↗
03 / 3D ASSET WORKFLOWBring 3D assets from Blender into an iPhone app.
We are defining a consistent workflow: prepare and check models and materials in Python and Blender, then export USDZ assets for Apple RealityKit. The goal is to reduce repetitive cleanup and catch issues before the assets reach the app.
Python/Blender→Prepare models, materials, and specs→USDZ/RealityKit
Progress: shaping a consistent process for creating and importing assets.
04 / EVO GO · IOSMany AI capabilities, brought together on one phone.
evo go envisions specialized small AI models running entirely on iPhone, sharing each task and bringing their work together through one assistant. Describe what you need, then decide what to do with the result.
We hope collaboration between small models can deliver complex-task performance comparable to cloud-based large models. Actual performance still needs to be tested on a phone.
One mobile assistantSmall models working togetherAll-on-device goal
Progress: defining the product architecture and capability scope.
Explore evo go ↗
05 / APPLIED AI RESEARCHWhat does it take for AI to complete a task well?
We study whether AI can handle tasks with several steps, let different models take on the parts they do best, and put the results to work in real business settings. We also assess deployment needs, oversight, and potential uses.
We evaluate each study on its own. Controlled tests do not prove that AI can work reliably over time, improve itself, or deliver business results.
Tasks with several stepsModels sharing the workReal-world use
Progress: testing how AI handles tasks and what it would take to use it in practice.
Explore research topics ↗