InTune Get in touch
Validation

What we are trying to find out.

InTune is a hypothesis, not a conclusion. This page keeps the open questions, the experiments and the findings in one place — including the ones that are still unanswered.

Questions

Six questions the project has to answer.

Product

Do people actually value this type of human support?

Supporters

Can appropriately trained non-licensed Supporters provide useful support within clearly defined boundaries?

AI

Does AI actually make the Supporter better?

UX

Can AI provide assistance without distracting from the human conversation?

Safety

Can the system identify situations that require escalation?

Economics

Can the model eventually operate at a sustainable cost?

Validation so far

Professional feedback.

The concept has been discussed with psychology professionals and researchers, who were asked to challenge it rather than approve it. The feedback was sufficiently positive to continue developing and testing the concept — see Research.

Next stage

A controlled pilot.

The pilot should test the system in a controlled environment, with defined measurements in four areas:

Supporter experience

Did the AI help? Was it distracting? Did it improve confidence? Did it save cognitive effort?

Session quality

Did conversations improve? Did Supporters ask better questions? Did they miss fewer important details?

AI quality

Accuracy, relevance, timing, hallucinations, false positives, false negatives.

Safety

Risk identification, escalation quality, boundary violations.

Experiments

Every experiment recorded the same way.

Question · hypothesis · method · result · conclusion · decision. The format matters more than any single experiment: it makes it possible to see later why something changed.

Experiment 001To test

Should the AI proactively interrupt the Supporter during a conversation?

Hypothesis: frequent interruptions may increase cognitive load and interfere with natural conversation.

Method: compare proactive suggestions against on-demand AI assistance.

Result: pending.

Open questions

What is still undecided.

Rather than pretending everything has been settled, the uncertainty is kept visible.

Product

  • What is the exact boundary between Supporter and therapist?
  • How much training does a Supporter need?
  • How should Supporters be matched with clients?
  • How long should sessions be?

AI

  • When should AI speak proactively, and when stay silent?
  • How much context should be sent to the model?
  • How do we evaluate AI usefulness?
  • How do we measure hallucination risk?

Safety

  • What signals should trigger escalation?
  • Who makes the escalation decision?
  • What happens during a high-risk session?
  • What professional oversight is required?

Business & technology

  • Who pays, and what should Supporters earn?
  • Is $5 per session economically realistic?
  • Which market should be piloted first?
  • Which transcription provider, and how are transcripts stored?
  • What should be real-time versus post-session?
  • What happens when the AI or transcription service fails?