Technology

Two engines inside every plan.

µLayer builds each dish the way a home kitchen does, layer by layer. µSolve chooses the dishes for every meal and sizes every gram against up to 43 ICMR-NIN 2020 targets, inside every limit. Neither is a language model, so every number in a plan traces back to a food and a rule.1

  • The plan engine · say “micro-solve”

    µSolve™ 1.0

    Solves each day as a mathematical programme: one plate per meal, every gram sized, every upper limit kept. What a day misses carries into the next, so the week balances.

    How µSolve works

  • The recipe engine · say “micro-layer”

    µLayer™ 1.0

    Builds 449 Andhra and pan-Indian dishes in five layers: a protein core, support and outer grain, fibre and helpers, each part with the range a cook would use.

    How µLayer works

  • Nutrient targets solved for

    43

    From ICMR-NIN 2020, with floors and ceilings, for the client’s own group.

  • Foods in the table

    577

    IFCT 2017, with a few staples from USDA, each with its cooking losses.

  • Dishes, built in five layers

    449

    269 Andhra, named in Telugu and English, and 180 pan-Indian.

  • AI credits per plan

    0

    Plans are solved, not written by a language model, so there is no meter.

µSolve 1.0 · the plan engine

It solves for the numbers. It doesn’t write them.

AI meal planners write a menu and estimate its nutrition as they go. µSolve works the other way round. It starts from the client’s targets and limits and searches real plates for the grams that come closest, so the same food always gives the same numbers.

  1. Step 1

    Targets

    A floor, a goal and a ceiling for every nutrient, for this client.

  2. Step 2

    Candidates

    Only the plates a home would serve, that fit the client.

  3. Step 3

    The solve

    Each day as one problem: plates chosen, grams sized, limits kept.

  4. Step 4

    The week

    Each day’s shortfall carried into the next, so the week balances.

  5. Step 5

    The checks

    Every number worked out again from the final grams.

Step 1 · Targets

A floor, a goal and a ceiling.

Energy comes from the Mifflin-St Jeor equation and the client’s activity, or from Katch-McArdle when you enter body fat. Every nutrient then gets ICMR-NIN 2020’s floor (the EAR), goal (the RDA or adequate intake) and ceiling (the tolerable upper limit) for the client’s age, sex, activity, pregnancy or breastfeeding. Conditions, blood work and medicines move a goal or add a cap: 25 condition protocols, 27 blood-work rules and 23 drug-food rules. Your own prescription always wins.

Iron for a woman of 30, from ICMR-NIN 2020. Anaemia doubles the goal, but a raised goal stops at 90% of the upper limit, so 58 mg becomes 40.5 mg. The ceiling never moves.

Step 2 · Candidates

Only plates a home would serve.

µSolve never invents a dish. It works from µLayer’s plates: dishes that fit the client’s diet, allergies, exclusions and your rules, put together the way their cuisine serves them. Each day it draws a fresh set of candidates, weighted towards what the day still needs and away from what the week has already had.

Candidate plates are drawn again each day, weighted towards dishes rich in what the day still needs and away from what the week has already had.

Step 3 · The solve

Each day, solved as one problem.

Each day is a mixed-integer linear programme. µSolve decides which plate each meal gets and how many grams of each ingredient go on it, keeps every hard limit, and makes the weighted shortfalls as small as the food allows. When no plates can keep a limit, the day is flagged for your review with the reason; it is never quietly loosened.2

A mixed-integer linear programme, solved with the open-source CBC solver through PuLP. Food values and targets are fixed tables, so the same food always counts the same.

Step 4 · The week

Seven days, balanced as a week.

ICMR-NIN 2020 sets vitamin and mineral needs over habitual intake, so µSolve judges them on the week’s average. Days are solved in order, and a day that falls short raises the next day’s target by 60% of the gap, never more than 35% above the goal. Main dishes don’t repeat within seven days.

Illustrative calcium, target 1,000 mg. Monday falls 180 mg short, so Tuesday's target rises by 60% of the gap, to 1,108 mg. The week averages 996 mg. Energy, carbohydrate, fat and sodium are held to each day's own target instead.

Step 5 · The checks

Then every number is worked out again.

Before a plan reaches you it passes a final gate that trusts nothing it was handed, not even the solver’s own totals.

  • From the grams up

    Amounts are rounded to kitchen steps of 5 g, 1 g or 0.5 g, and every nutrient is recomputed from them with cooking losses. Totals sent from a browser are never trusted.

  • Allergens and diet

    Every food is matched to a known identity and checked against allergies by every name, exclusions and diet type. A conflict stops the plan.

  • Medicines

    23 drug-food rules across 20 medicines, from warfarin’s steady vitamin K to metformin’s B₁₂, checked across every day.

  • Limits, again

    Upper limits, clinical caps, added sugar and absorption are checked once more, and anything outside them is reported.

  • The gap report

    Each nutrient’s week against its floor and goal, with the reason when food alone can’t reach it, such as vitamin D.

  • On the record

    Each plan records its seed, the version of its target table and its data sources, so you can see what it was built from.

µLayer 1.0 · the recipe engine

Every dish, built in five layers.

A recipe database stores finished recipes at fixed amounts. µLayer builds each of 449 dishes the way a home kitchen does, layer by layer: a protein core, the grain, the vegetables and the helpers that season them. Each layer has its own ranges and its own job, so µSolve can change 45 g of dal to 52 g and the method still reads right.

Two meals from real one-day µSolve plans. Raw weights for one serving. A layer a meal does not need is left out: rice is the outer grain, wheat a supporting one.

  • Layers with a job

    When you prescribe more protein than usual, µSolve lets only the protein core grow, up to 1.8 times its range, within each meal’s portion limits. Oil stays a helper, inside each recipe’s small range.

  • Cooked, not raw

    Every dish family maps to a cooking method, and every food to one of 270 USDA retention factors, so a plan counts what is left after cooking. Added salt is measured per dish.

  • Measures a home uses

    Katoris, spoons and pieces from the household-measure tables of the Dietary Guidelines for Indians 2024, beside every gram you see.

Every meal in Pro lists its layers, with the protein each one gives, under Recipe layers.

Inside a dish

Thotakura pappu, layer by layer.

The dal is the protein core, the amaranth is the fibre, and the oil, tempering and souring are helpers. µSolve sizes the three flexible parts inside the ranges a real cook would use. The method is a template that takes whatever amounts µSolve chooses, so the grams in the steps always match the grams in the plan.

One standard serving: the middle of each range. In a plan, µSolve chooses the three amounts for the client and the method is rewritten with them.

Inside a meal

A plate has rules too.

µLayer knows how a cuisine serves dishes together: what goes beside a pesarattu, when curd finishes a rice meal, and which pairings an Andhra home would never make. µSolve only ever chooses between plates that follow these rules. You set how often a pachadi, chutney or podi comes with each meal.

The Andhra pack's own rules, from sourced home-meal patterns. Pan-Indian plates follow the pan-Indian pack's rules.

Your rules

Must, Prefer, Avoid, Never.

Tell both engines what this client’s week needs, on a food, a dish or one part of a meal, for the meals and days you choose. Up to 40 rules a client, and every plan reports how it kept each one.

  • Must

    At least once in the meals you pick, or a number of times a week, spread evenly. A must outranks the variety rule.

  • Never

    Never, at most a number of times, or a gram cap per meal. Never beats Must for the same meal, and the plan names the clash.

  • Avoid

    Kept out unless a day’s targets can’t be met without it. The plan says which meals, if any, it had to use.

  • Prefer

    Drawn more often and costed lower, so it appears wherever it fits.

On every paid plan

Both engines, from Basic.

Every paid plan builds with µSolve and µLayer, with your clinic’s name on the result. What changes is how many targets µSolve works to reach. The limits never change: the same client gets the same safety checks on every plan, Free included.

Free shows Day 1 of each plan, balanced like Premium. Every plan reports all 48 nutrients and names every gap; a plan only changes which targets the solver works to reach.
Compare plans

Where AI is used

No language model picks a food or a gram.

µSolve and µLayer are arithmetic over published tables. Where you switch it on, a language model (Google Gemini) drafts the plain-language “Why this plan” note and a patient brief. You read it before your client does, and it never changes a number.

Read the methods and sources

What it doesn’t do

Honest about its edges.

  • A draft for your review

    Every plan is a draft for a qualified professional to check, adjust and approve. It supports your judgement; it is not medical advice.

  • Two cuisines today

    Andhra Pradesh and pan-Indian packs are live, with a blended mode. Tamil Nadu is written and in review; other regions plan from the pan-Indian pack.

  • Estimated cooked yields

    Katori sizes and cooked weights come from published factors. They have not yet been weighed in a test kitchen.

  • What food can’t reach

    Vitamin D, and B₁₂ on a vegetarian week, are often short from food alone. The gap report says so instead of rounding up.

Watch it solve a real week.

Start free with your own clients, no card needed. Or let us show you in a 20-minute demo.

Book a demo