BuildOrbit
All Projects

Paediatric Growth AI

A growth consult for paediatricians: dictate or scan a child’s data, and get the centile chart, red flags, a clinical summary and a diet plan.

A working demo built for an AI learning course for paediatricians. The doctor dictates in Hinglish, scans a handwritten OPD card or types, and the app plots the child on WHO 2006 or IAP 2015 growth charts with rule-based red flags. AI then drafts a clinical summary and a seven-day Indian diet plan around those calculated facts, and a Learn mode explains each step.

Client delivery. The product was designed and engineered as studio work for the team that owns it.

Related serviceAI integration for SaaS
HealthcareAI Clinical Teaching DemoWorking demoWeb
Discuss a Similar Project

What the demo does.

A pharma client was running an AI learning course for paediatricians and wanted a live demo for it. It had to feel like a real consult, run on venue Wi-Fi, and teach one clear idea: AI can take the typing out of a consult, but the clinical numbers must be calculated, not generated.

  • Dictate in Hinglish

    The doctor speaks the child’s details the way they would in clinic, and the values land in the right fields.

  • Read an old OPD card

    A photo of a handwritten card becomes past visits on the child’s growth timeline.

  • Plot the right chart

    WHO 2006 under five years, IAP 2015 from five to eighteen, with z-scores and growth velocity.

  • Flag what needs attention

    Stunting, crossing centile lines and falling below the family target are rule-based, not guessed.

  • Draft the summary and diet plan

    A clinical summary to edit and approve, and a seven-day Indian diet plan to ICMR-NIN targets.

  • Teach how it works

    Learn mode explains the AI behind each step, and case challenges turn it into course material.

Capture

Say it, scan it or type it. One form either way.

Voice and scan are shortcuts that fill the same form. The doctor dictates the child’s name, age, height, weight, parents’ heights and diet in Hinglish, and can scan an old OPD card whose rows are written by age. Every value stays editable before anything is plotted.

  • Hinglish transcription into structured fields
  • Vision AI reads handwritten OPD-card rows
  • Manual entry with checks on impossible values
Capture step: a Hinglish dictation transcript beside the filled patient details form and three past visits
Recreated from the product, with sample data.

The chart

Percentiles are calculated, not generated.

The chart, the centile, the z-score and the growth velocity come from the LMS method on WHO 2006 and IAP 2015 reference data. The IAP values are fitted from the published centile tables. Red flags are rules on those numbers, so the same child always gets the same answer.

  • LMS method on WHO 2006 and IAP 2015 data
  • Measured value, centile, z-score and velocity
  • Rule-based red flags, never model output
Growth assessment chart with the child's points below the 3rd centile, a z-score of -2.60 and red flags for stunting and crossing centile lines
Recreated from the product, with sample data.

Summary and plan

AI drafts the words around the facts.

Only after the maths is done does AI write. It receives the calculated facts and is told not to change them, then streams a short clinical summary for the doctor to edit and approve, and a seven-day regional diet plan checked against ICMR-NIN 2020 targets. No patient names are sent to the reasoning model.

  • Summary is a draft for the doctor’s review
  • Diet plan rings compare the week with targets
  • Patient names stay out of the reasoning prompt
Clinical summary noting the child has crossed down centiles, beside a seven-day Indian diet plan and follow-up flags
Recreated from the product, with sample data.

For the course

Built to be taught, and to survive the venue.

Learn mode walks through the AI behind each step in plain clinical language, and case challenges ask doctors to predict before the reveal. Three sample patients have cached fallbacks, so the demo still runs if the venue Wi-Fi drops.

  • Learn mode in plain clinical language
  • Case challenges: predict, then reveal
  • Offline fallbacks for the sample patients
Recreated from the product, with sample data.
For the technical folksHow it’s builtThe full feature list and the tools behind it.
  • Hinglish voice dictation into structured patient fields
  • Handwritten OPD-card reading with vision AI, dated from the child’s age
  • WHO 2006 (0 to 5 years) and IAP 2015 (5 to 18 years) centile charts
  • Rule-based red flags for stunting, crossing centile lines and family target
  • Streamed clinical summary for the doctor to edit and approve
  • Seven-day Indian diet plan to ICMR-NIN 2020 targets
  • Learn mode and case challenges for course teaching
  • No database: patient data stays in the browser tab
Next.jsReactTypeScriptOpenAIZustandTailwind CSSPlaywrightVitest

Explore adjacent case studies.

If this build is close to what you have in mind, these projects show the range around it.

More case studiesDiabetes Distress Assessment Patient Assessment PlatformHandwriting Insight AI AI Engagement ExperienceAwareness Pledge Awareness Campaign Platform

Building something with this level of complexity?

If you're planning a ai clinical teaching demo with real delivery constraints, we can help you scope it, make the technical calls early, and ship it properly.

Related serviceAI integration for SaaS
Start a Project
Direct technical conversations before scope gets expensive.
Full-stack product delivery from interface to backend to launch.
Clear tradeoffs and execution plans instead of vague agency promises.