Machine Learning Intern, Neural Care

What the company does

Neural Care builds AI-powered support for people caring for someone with dementia.

What I do

My job is evaluation: working out whether the AI agents behave the way a caregiver actually needs them to.

• Ran five meetings with subject-matter experts to map the caregiving workflows the agents carry out, rather than guessing at them from outside.
• Prepared an evaluation plan for those workflows and presented it.
• Revised it three times on expert feedback, and won leadership approval for the final document.
• Now implementing that plan on open-source models, five hours a week at weekends.

Why this and not something else

My research is about refusing to take a model’s answer on trust. This is the same question asked of a product rather than a paper: how would anyone know whether it is working?