MIRA
Pain physicians start each visit with an AI summary of the patient's history
MIRA's founder, a practicing pain clinician, wanted AI to take each history before the visit. We designed and built adaptive intake answered by tap or voice, with an AI summary the physician reads, and it ran in three practices.
Healthcare & WellnessAI patient intakeBuilt from zero
The challenge
A pain clinician wanted AI to take each history before the visit: pain, past medical and surgical history, medications with doses and intervals, and treatments already tried.
What we did
We designed adaptive intake with the founder, with required answers, doses as choices and voice answers, plus an AI summary for the physician, live on Azure in three practices.
The outcome
The founder approved the first version after running it with patients in practice, and what it showed set the brief for the multi-clinic product.
- Built for
- The founder's own pain practice, then two more
- Practices
- Three, each on its own Azure deployment
- Covers
- Pain, history, medications with doses, treatments tried
- Answers
- Tap, yes or no, condition lists or voice
The starting point
MIRA is AI patient intake founded by a practicing clinician who treats pain patients. Patients answer before they see the physician, and the physician starts the visit with a structured summary of their history.
What had to change
A full pain history covers pain, past medical and surgical history, current and past medications with doses and intervals, and treatments already tried. The founder wanted AI to gather it before the visit and hand the physician a summary complete enough to use.
What could not break
- The AI drafts a summary for the physician and never diagnoses
- Required answers, so patients cannot click through and leave gaps
- Medication detail includes dose and interval, not only the drug name
- No patient names or birth dates stored, only a generic ID per intake
- HTTPS on the front and back end
The calls we made
We treated the AI as a careful resident taking the history for the physician: thorough questions in, a draft summary out, and every clinical call left to the physician.
- 01
Adaptive, with required answers
Follow-up questions change with each patient's answers, and required answers stop patients from clicking through, so the summary is not built on gaps.
- 02
Doses and intervals as choices
Current and past medications come with common doses and intervals to pick from, so drug, dose and interval reach the physician together and ready to use.
- 03
Voice as well as tap
Patients can speak an answer, transcribed by OpenAI Whisper into the same flow, alongside yes or no questions and condition lists.
- 04
AI drafts, the physician decides
The model writes a structured summary for the physician to read before the visit. It does not diagnose, and the physician leads the visit.
- 05
Prove it before storing identity
Each intake got a generic unique ID instead of the patient's name or date of birth, so the first version could run with real patients without holding their identity.
How the system fits together
What we built
Adaptive intake flow
Pain assessment, past medical and surgical history, current and past medications and treatments tried, with follow-up questions that change with each answer and a question order refined with the founder.
Voice answers
Patients can answer by voice as well as by tapping yes or no or picking conditions from a list, and OpenAI Whisper transcribes each spoken answer into the same flow.
AI clinical summary
OpenAI GPT turns the answers into a structured summary that the physician reads in a physician view before the visit, without a diagnosis.
Required answers
Key questions need an answer before the patient moves on, so nobody clicks through and leaves the physician a summary with gaps.
Medication detail
Current and past medications come with common doses and intervals as choices, so drug, dose and interval reach the physician together.
How it works
Answer
Before the visit, the patient answers the intake questions by tapping or speaking.
Adapt
Follow-up questions change with each answer, and required ones cannot be skipped.
Transcribe
OpenAI Whisper turns spoken answers into text inside the same flow.
Summarize
OpenAI GPT drafts a structured summary of the pain, history, medications and treatments tried.
Review
The physician reads the summary and leads the visit.
A closer look
Intake summary
Intake ID: 7F3K-2291
Chief complaint: Lower back pain spreading to the left leg
Pain assessment: Constant ache with sharp flares, worse when
sitting, eased by lying down
Past medical history: Type 2 diabetes, high blood pressure
Past surgical history: Right knee arthroscopy
Current medications:
- Gabapentin 300 mg, three times a day
- Metformin 500 mg, twice a day
Past medications:
- Tramadol 50 mg, as needed, stopped
Treatments tried: Physical therapy, chiropractic care,
one epidural steroid injectionWhat changed
Physicians started visits with the patient's pain, history, medications with doses and treatments tried already summarized. The founder approved the first version, three practices ran it, and it led straight to the multi-clinic product.
“Amazing level of proficiency and responsiveness. We are looking forward to moving on to our next project!”
Under the hood
- Datics role
- Product built by Datics
- AI
- OpenAI GPT writes the clinical summary from the patient's answers, and OpenAI Whisper turns spoken answers into text.
- Patient data
- No patient identifiers stored in the first version, only a generic unique ID per intake, with HTTPS on the front and back end.
- Hosting
- Microsoft Azure, with three separate deployments on three domains, one for each practice.
- Boundary
- The summary is a draft for the physician to read before the visit. The AI does not diagnose.
AI and models
Cloud and infrastructure
Apps and APIs
How it shipped
Design with the clinician
We designed the intake with the founder around the history a pain physician needs: pain, past medical and surgical history, medications and treatments tried.
Build and refine the flow
Pain assessment, medications with doses and intervals, required answers and question order, refined with the founder, plus voice answers and the AI summary.
Live in three practices
Deployed on Microsoft Azure as three separate deployments on three domains, with HTTPS and no patient identifiers stored.
Approval and the next phase
The founder approved the first version, and what it showed in practice set the brief for the multi-clinic product.
Built with the clinician
We reviewed each flow with the founder, who ran it with patients in the founder's own practice and told us what physicians needed from the summary.
Live in practices early
The first version ran in three practices on Microsoft Azure, each with its own domain and environment, so feedback came from real visits.
What this unlocked
Its use in the founder's practice led straight to phase two: one product any clinic can sign up for, pay for per provider and run on its own.
Related work
Have a product with the same kind of constraints?
Tell us what already exists and what you want it to do. We will tell you what it takes.