Methodology

How Miriel actually works.

No black box, no marketing magic. Miriel produces every recommendation by combining pediatric clinical standards with AI models, in a specific order, with safety always on top. Here is what that means in practice.

The shape of every recommendation

Every meal plan, food scan, and AI coach answer passes through the same evaluation, in the same order:

Safety first. Allergens, age-appropriate choking-hazard rules, and any medical-condition exclusions on the child profile are evaluated as hard constraints.
Pediatric standards next. Growth percentile, age-tuned nutrient and portion targets, and stage-appropriate texture and energy needs are checked against current WHO, CDC, AAP, and Korean pediatric guidance.
Behaviour and acceptance. Picky-eating profile, food jags, and what the child has actually eaten before are taken into account so plans are realistic, not idealised.
Family context. Dietary preferences, cuisine, what is in your pantry, sibling profiles, and (when shared) school meals are layered in.
Preference and variety last. Within whatever space the previous layers leave, Miriel optimises for variety, parental cooking effort, and child enjoyment.

The order is not negotiable. A more-fun meal plan does not override an allergen warning; a child's preference does not override the calorie ceiling for their age. Safety beats personalisation, every time.

The capabilities behind that order

Miriel is not a single AI model with one prompt. It is a set of pediatric nutrition capabilities, each tuned to one part of the problem. We group them at a high level here; the specific implementation evolves with research and is not disclosed in detail.

Growth and energy

Miriel interprets a child's growth percentile against age- and sex-appropriate standards, projects expected energy and protein needs for the next growth window, and adjusts recommendations for catch-up growth or growth spurts. Adult calorie logic is not scaled down — pediatric growth has its own targets.

Nutrient balance

Macronutrient and micronutrient targets are tracked against intake patterns over days and weeks. Common pediatric deficiency risks — iron, vitamin D, calcium, B12, zinc, omega-3 — are watched specifically rather than averaged into a generic "balanced" score.

Allergen and condition safety

Parent-declared allergies, intolerances, and medical conditions (celiac, EoE, FPIES, CMPA, and more) are first-class data. They cannot be over-ridden by preference, cuisine, or pantry availability. Cross-contamination warnings surface on packaged-food scans.

Choking and developmental appropriateness

Food shape, texture, and size are filtered by age. Whole grapes, whole nuts, and other high-risk items are flagged or modified for under-4s without parents needing to remember the list themselves.

Picky eating and acceptance

Miriel tracks what your child has actually eaten and accepted. It uses repeated-exposure principles (Birch & Marlin, Satter) to expand the diet gradually rather than forcing unfamiliar foods. Picky-eater profiles are explicitly supported, not treated as edge cases.

Cultural cuisine and dietary patterns

Korean, Western, vegetarian, vegan, halal, and kosher patterns are supported. Cuisine is treated as a real constraint, not a coat of paint — Korean meal plans look like Korean meals, not Western meals translated into Korean.

Family context

Multi-child households, sibling differences, pantry contents, school lunch reconciliation, and cooking-skill realism all feed into the recommendation. A single meal can be planned across siblings with different needs without becoming a separate-meals nightmare for the parent.

How a meal plan actually gets generated

Parent indicates which child the plan is for and the time window (a single meal, a day, a week).
Miriel pulls the child's safety profile (allergies, age, medical conditions, choking-hazard rules).
Pediatric energy and nutrient targets for that window are computed from growth standards and DRI/KDRI.
Miriel consults what the child has accepted in the past, what is in the family's pantry, and the family's cuisine and dietary preferences.
The AI models propose a plan that fits all of the above. The plan is checked against the safety profile a second time; nothing leaves Miriel that violates a safety constraint.
The plan is returned with allergen flags, age-appropriate texture notes, and (where relevant) prep instructions.

What Miriel does not do

— Miriel does not diagnose medical conditions, allergies, or feeding disorders.
— Miriel does not replace a pediatrician, registered dietitian, or feeding therapist.
— Miriel does not provide medical or clinical advice. Recommendations are educational and operational.
— Miriel does not generate calorie-deficit weight-loss plans for children. Pediatric weight concerns are managed by a pediatrician, not by an app.
— Miriel does not adapt to non-evidence-based diet fads. We follow current pediatric and public-health consensus.

Data sources and model choices

Pediatric standards are summarised on the research page. Food data — labels, ingredients, nutrition panels — comes from a combination of regulated label data and curated databases, with on-device scanning that reads the front of the packet you point the camera at.

The AI side uses frontier multimodal models from leading providers. The specific model choices, prompt designs, and internal weighting are not published in full — they evolve with research and are part of how Miriel differs from generic adult-nutrition apps. What is committed in writing is the order above, the references on the research page, and the safety-first rule that overrides everything else.

Honest transparency limits

We do not publish the specific algorithms or internal weighting behind individual recommendations. There are two reasons. First, the implementation is part of Miriel's competitive surface — it is what we have built that adult-nutrition apps rescaled for children do not have. Second, narrowly-published mechanisms invite reverse-engineering that pretends to honour safety while quietly working around it.

For pediatricians, dietitians, schools, or researchers who would like a detailed walk-through under NDA — for validation, partnership, or clinical evaluation purposes — we are happy to do that directly. The contact form below reaches the right people.