Health AI SaaS · 10 / 11 · web application
Nutritionist Webapp
The console nutritionists use to review a member's day, built around the worst meal: why it scored badly, and what each fix would be worth.
Overview
This is the console Twin Health's nutritionists and health coaches use to review a member's day: what they ate, what each meal did to their glucose, the symptoms they reported, and the five-day meal plan that comes next. A nutritionist works through a long list of members with it every day.
The problem: Reviewing one member meant cross-referencing glucose spikes against food logs by hand, finding the food that caused them, and writing a meal plan that fitted the member's tastes and culture — upwards of 45 minutes each, which capped how many members a nutritionist could support.
What we built: A daily report built around the worst meal. One action score sums up the day; each meal has its own score; the lowest one says what in it did the damage and what would undo it. Alternatives are scored before they're suggested, so the coach sees the size of a change before making it. Symptoms are switches with a remedy each, and a five-day plan is written from the member's own likes, ready to review rather than start from scratch.
Who it's for:
- Primarily: clinical nutritionists, registered dietitians and metabolic health coaches.
- Also: diabetologists checking a member's diet against their medicines.

Good
11 of 12 tasks done
150
mg/dL120/80
mmHg65
kg4,440
todayLooks like the boiled egg at dinner is pulling your metabolic score down. A 20-minute walk after the meal would even it out.
Find the worst meal, size the fix, send it
In the prototype above, pick Dinner, apply an alternative and watch both scores move. Then write a plan.
The questions it had to answer quickly:
- How was their day? In under ten seconds: glucose, blood pressure, weight, steps and meals, together.
- Which food did it? Tie a post-dinner spike to the thing on the plate that caused it.
- What should change? Offer a fix with its effect attached, not a lecture.
- What next? A meal plan the member will actually eat, without writing one from nothing.

The session header
A member switcher (12782 — Rajan Elangovan), a date with back and forward, Cancel and Save, and tabs for sensors, supplements and medicines, self-reports and the daily report.
The action score
A ring — 93, great — with 11 of 12 tasks done and four tiles: glucose 150 mg/dL, blood pressure 120/80, weight 65 kg, 4,440 steps. Underneath, a nudge for the member: "You're one step away from 100 three days in a row."
Meal by meal

Four rings in time order — breakfast 90 at 9:30, lunch 72 at 1:30, snack 43 at 5:30, dinner 27 at 8:30 — and for the low ones, the reason and a fix: "Taking boiled egg for dinner is affecting your metabolic score. A 20-minute walk after the meal would help." Alternatives lists the other options, each with what it would add.
Symptoms and remedies
Reported symptoms are switches — Headache, mild; Night urination, high — and switching one on brings up the remedy: "Avoid citrus drinks such as lemonade, orange and grapefruit juice to ease night-time urination," with an article to send.

The meal plan
Five days (15–20 August) written from four things the console shows plainly — the member's preferences, members with a similar taste profile, food ratings, and the meal plan engine — with Review plan to check it. A queue of foods waiting for the member's rating, and a drawer of short articles (How Twin nutrition differs from keto or paleo) to share.

Results
- Caseload: from 25 members per nutritionist a day to 95 — 3.8 times as many.
- Meal plans: a five-day plan went from about 35 minutes to under 4.
- Evenings: members on coach-curated plans had about one in five fewer evening glucose spikes over a month.
Conclusion
The console didn't make nutritionists faster at the old job — reading logs and searching for causes. It did that part for them, so their time went on the judgement: which fix to suggest, how to say it, and whether the plan fits the person.
Gallery
01 / 25The console — Member, date and the action score at the top of the day.



















