Selected ProjectExpertise & Other ProjectsAbout meContact

Capturing the lived experience of recovery

A research platform that captures what trial participants go through between check-ins. An AI agent checks in with them by voice, and a team of agents helps researchers turn those conversations into qualitative data for their study.

Role
Design Engineer
Project owner
Finnish Centre for Evidence-Based Orthopaedics (FICEBO), Helsinki University Hospital
Timeline
January 2026 to now
Status
MVP v2 in development, grant application in process
Feedback
“I only asked for one talking agent. I never imagined you would build a whole working platform.”
Project owner, FICEBO
“This is a whole new way of doing medical research.”
Senior researcher, FICEBO

Platform Introduction

The platform has two sides. Participants talk to an AI agent about their health on their phone, and researchers define the trial, monitor and control the agents in the system, and get transcripts, patterns and help in creating new medical knowledge.

Participant app
Researcher workspace
Demo
Demo video
The Platform’s Two Sides

Problem Brief

Two patients with the same injury and the same treatment can recover completely differently. Recovery is very much affected by a person’s lifestyle, experiences and environment.

Check-ins are infrequent, with long gaps in between, and the lived experiences of trial participants are not collected. Participants are asked to score their feelings and pain levels, and many are confused about how to do that.

Medical research as it isWhat it misses, and what could be there

Capturing that experience means interviewing participants regularly, then transcribing, pseudonymising and analysing every conversation manually, at a scale no research group can afford. So in MSK research it mostly isn’t done.

Investigating the Problem and Opportunity

We explored the problem and the opportunity space in depth through interviews with patients, research participants, industry experts and researchers. We made field visits to the hospital, and to Oxford in the UK and Amsterdam in the Netherlands, meeting experts there and bringing in their points of view.

My Role

Three of us from the capstone team continued over the summer: a service designer and UX researcher, a UX, UI and visual designer, and me, coming from the School of Engineering, as the design engineer. The role was more than building the platform. It meant thinking on many different layers and taking on many different tasks.

Designing the ProductThe participant app and the researcher workspace, and an experience through which the functions can actually be used.
Developing and TestingI develop and test it in steps: feasibility first, aesthetics later.
Bringing Stakeholders AlongConvincing the research group at HUS, getting them interested, and bringing in collaborators from outside the group.
Navigating RegulationWorking with my colleague on what the research permit allows, and meeting the HUS IT department to define what the platform needs before it can run in clinical trials.

How the Product Was Developed in Phases

Examples of Problem Solving and Critical Thinking

Human practice and thinking →agentic system and architecture

I take a human way of thinking or working and convert it into an agentic system or architecture. These are examples. I came across many more problems.

01

The platform needs to be research-question centric

ObservationIn my own research and user studies, I need to know what I am looking for before deciding what to collect. The same goes for the platform we are building. A patient-centric platform that collects data in general, and analyses it much later for insights, would not lead to results.
ImplementationWithin our platform, a trial cannot be created without its research question and goals. They are the upstream point, and they inform the behaviour of every agent downstream.
02

Adopting the interaction model of a nurse or doctor calling a research participant

ObservationThe quality and frequency of data collection depend on who has to take time out to record the information. If a nurse calls to check in, they carry the responsibility and the participant only needs to answer. A diary is different: the participant has to take the initiative.
ImplementationThe check-in agent makes the call, so the participant can be passive.
03

How a human interviewer remembers a participant

ObservationWhen I conduct interviews, the last session is usually sharp in my mind, and I remember a few key points from earlier sessions. I can also review the participant’s profile before each interview.
ImplementationTaking inspiration from the moving average (a mathematical concept), I developed a rolling memory. It is updated after each session by a dedicated agent, the Memorizer.
04

Working within regulatory limits

Observation from interviewsWho is responsible for the participant’s health? Do researchers give medical advice during a study? What are an interviewer’s responsibilities, and what should they do in critical situations? And what do the regulations on medical devices require?
ImplementationWe applied safety limits to the conversational agent so that it never gives medical advice, even by mistake. In critical situations it can flag concerns to the relevant research nurse so that further steps can be taken, but it does not take responsibility for them.

User Testing

In July, ten healthy volunteers used the platform for a week on their own phones, with no one supervising. We could not use real health data yet, so they talked about sports and health. They found it easy to use and talked more openly than in a survey, but felt they got little back. Speech with an accent was sometimes transcribed into the wrong language, so participants now choose Finnish or English before each call.

10volunteers
7days
34calls

Impact

From quantitative to mixed methods

MSK research relies heavily on quantitative studies. What we are building shifts it to mixed methods, which can bring in new information and fill in gaps. In the long run, it is about delivering precise care to patients depending on their own situations.

Next: Preparing to run in actual clinical trials.

FICEBO on the project
FFICEBO · Article How Can We Answer the Questions Patients Really Want Answered? Why trial questionnaires miss what patients go through, and how an AI interviewer, built with Aalto University students and tested with ten volunteers, could fill that gap. Read on ficebo.com FFICEBOFinnish Centre for Evidence-Based Orthopaedicsin An AI-based project exploring new ways to collect valuable data from patients. Watch the video on LinkedIn View the post on LinkedIn