Inside Kelbetty: turning vague pain into a clear note for the doctor

Most people are bad at describing pain. It's not that they're careless; pain is subjective, and the words for it are slippery. “It kind of burns, but also aches, mostly in the morning, I think?” By the time a patient finds the words, half the appointment is gone. That gap is what Kelbetty exists to close.

Kelbetty is a symptom-to-summary tool. A patient answers a series of simple, guided questions, and the system turns their answers into a clear, structured summary to bring to a doctor: written in plain language for the patient, and in medical terms for the clinician. The tagline we settled on says it plainly: translating patient pain into clinical insight.

The hard part was never the model

When people hear “AI healthcare,” they picture the model doing something clever and clinical. The model is the easy part. The hard part is turning a vague, emotional, half-remembered description into something structured enough to be useful, without putting words in the patient's mouth. That's a product problem before it's a model problem, and it's where almost all of our work goes.

Our job isn't to sound like a doctor. It's to help a patient be understood by one.

How it actually works

There are three steps, and we fought to keep it that simple. First, the patient answers guided questions, with no medical knowledge required. The questionnaire is adaptive: it asks context-aware follow-ups, so “it burns” leads to questions about onset, location, triggers, and how it has changed over time, with a 0–10 scale and a body map to point at exactly where it hurts. Second, the system structures those answers into organized fields: main complaint, area, duration, progression, and any red flags. Third, the patient gets a summary they can keep or share right away, as a PDF or a link.

Why it's a conversation, not a form

The easy version of this product is a static form: twenty fields, submit, done. We deliberately didn't build that. A form asks everyone the same questions and learns nothing; an adaptive flow asks the next question because of the last answer, which is how a good clinician actually takes a history. It's more work to build and to tune, but it's the difference between a summary that's technically complete and one that's actually useful. When a caregiver fills it in for a child or an aging parent, the same logic switches to age-tuned prompts.

The line we won't cross

Kelbetty does not diagnose, does not prescribe, and is not a replacement for a doctor. That isn't a legal footnote we tacked on; it's a product decision that shapes everything. The moment a tool like this starts telling you what's wrong, two things happen at once: patients trust it in situations where they shouldn't, and it quietly becomes a regulated medical device. We made the deliberate choice to stop at structuring the patient's own words and handing them to a clinician. The doctor still decides; we just make sure they start from signal instead of noise.

This matters more than it sounds. Regulators like the FDA and the EU's medical-device rules judge a tool by what it actually does, not by the disclaimer at the bottom. A “does not diagnose” label means nothing if the product is busy guessing diagnoses. So we keep the scope honest: organize, translate, flag the obvious red flags, and get out of the way.

Trust is the product

In healthcare, trust isn't a feature you add at the end; it is the product. Health data is encrypted in transit and at rest, the patient decides what's saved, exported, or shared, and the model isn't trained on anyone's private health information. We're explicit that AI is involved (most patients want to know), and we keep a human, the doctor, firmly in the loop. None of this shows up in a demo, but all of it decides whether anyone uses the thing a second time.

Building Kelbetty reinforced what I believe about AI products in general: the model is rarely the moat. The moat is the hundred unglamorous decisions around it: which question to ask next, where to stop, what to never claim, how to earn trust. Healthcare just makes that obvious, because here the cost of getting it wrong is real. We're still early, but the principle is set: help the patient be understood, and let the doctor do the rest.