Henrik and Caterine

Conversation
for learning and exploring.

How might we use situational awareness and conversation to better learn and explore.

The problem

Most AI waits for instruction. It reacts to what you type or say — but it doesn't perceive what's happening around it. A learning companion needs to understand context: what you are doing, what you are showing, what you are trying to express — and respond accordingly.

How it works

Talki watches, listens, and responds — building a picture of what's happening without waiting for a prompt.

  • Recognises familiar faces
  • Follows hands and gestures
  • Identifies objects in the environment
  • Listens and responds to speech

Work Log

We document the experiments on the way.

Tools for Meta's Codec Avatars Tools for Meta's Codec Avatars

Tools for Meta's Codec Avatars

Developed advanced avatar tools and pipelines at Meta Reality Labs, spanning from Instagram's stylized avatars to cutting-edge 3D Gaussian Splatting for VR experiences.
Role: Computer Vision , 3d Graphics , Research Engineering
Visual Evidence App Visual Evidence App

Visual Evidence App

Design and implementation for a white-label iOS App for creating camera driven workflows, where the user shows the system a situation, and the AI Agent determines what it is and what workflow is involved.
Role: Product Design , Mobile App , Ml Engineering , Full Stack