1 One-Line Concept
Tagline: Weigh once, raw or cooked. Scale anchors the weight. AI splits the nutrition.
Weigh once. The scale locks the weight. The AI splits the nutrition — with the right raw-vs-cooked data.
2 Problem Definition (Define)
Problem Statement
Fitness users need a faster, smarter way to track the nutrition of whatever they're eating — raw ingredients they're prepping or a meal they've already cooked, because weighing each item one by one and looking up calorie tables is both slow and error-prone — while pure-app guesswork isn't trusted.
How Might I
How might I help fitness enthusiasts accurately log the calories and macros of a whole portion — raw or cooked — with as few steps as possible?
What I'm NOT solving…
- Not a "weigh raw ingredients separately before cooking" workflow
- Not pure photo-based calorie estimation (no physical ground truth)
- Not a full fitness system that replaces training plans or recipe recommendations
What I AM solving…
- Trustworthy nutrition results for a whole portion — raw or cooked
- Uniting "accurate" and "effortless"
- A habit loop from daily logs → weekly reports
3 Target Users (Proto Personas)
Linda · 24 · Shanghai · Beginner
New to fitness, wants to cut fat; often copies routines from social media that don't fit her. Quick to abandon apps.
"I want to eat healthy, but I have no idea whether the meal I just made is actually balanced."
Product fit: learn nutrition from a real plate of food; ultra-simple steps; must clear the "is it accurate?" trust barrier.
Chen Hao · 31 · Beijing · Advanced
Trains regularly, tracks macros to build muscle; currently uses a basic kitchen scale plus notes on his phone.
"I know my targets — what I hate is spending ten-plus minutes calculating every meal."
Product fit: get every macro in a single action; eliminate the labor of weighing each item and looking up tables.
4 Key Decision Recap: Do I Still Need the Scale?
★ Decision: keep the scale. Actual weighing = accuracy reference (irreplaceable)
| Iteration | Approach | Conclusion |
|---|---|---|
| Iter 1 | Pure AI estimation: recognize → volume → m = ρ×V → sum | Simple hardware, but volume errors and density deviations compound, so the numbers drift |
| Iter 2 | Challenge: fitness users need trustworthy gram counts | No actual reading on the scale = no physical ground truth |
| Iter 3 ✓ | The scale measures the whole meal's actual total weight W + AI allocation | Main path Σ mᵢ = W_scale; the AI only handles relative shares |
Place the whole meal → the scale: actual total weight W (trusted — no debate over "is the scale accurate")
→ AI Sensor: food recognition + volume share; when quality is insufficient, Loop Engineering auto-retry (self-correction)
→ internally allocate mᵢ (Σ mᵢ = W) → look up the Calories & Elements table → app daily logs + weekly reports
Manual intervention only as an exception: food too dense, steam blocking the camera, etc. → resolve, then rescan
5 Concept Form · Keep / Reject
KEEP
- Covered bowl / closed-scan form · rear-hinge motorized lid
- AI visual recognition
- The scale base as accuracy reference
- Connected app (daily logs + weekly reports)
- Place the whole meal at once
REJECT
- Adding ingredients step by step, weighing each one
- Pure AI estimation with no actual weight reference
- Complex multi-step setup (drives beginner abandonment)
Best Idea
Whole-Portion AI Nutrition Scale — raw or cooked, via two snap-on bowls — AI vision sensing + motorized cover + precision scale base (actual total weight = accuracy reference) + app sync.
6 Six Key Features (scope entering Wireframe)
Weigh raw or cooked, whole
Snap on the raw or cooked bowl, set the whole portion down at once
Actual weighing (reference)
Lock the physical total weight W as the accuracy anchor
AI food recognition
Identify the food types on the plate (confirm / correct)
Volume + density allocation
Split the total weight across each food, Σ = W
Calories & elements table
Output kcal + protein / carbs / fat, etc.
App daily logs + weekly reports
Habit loop and insights
7 Research Evidence Summary (Concept Validity)
| Source | Insight | Impact on the concept |
|---|---|---|
| Beginner interviews | Abandon the app after two days; want to know "will this plate make me gain weight" | The flow must be very short; results must be readable at a glance |
| Advanced-user interviews | Weighing is still necessary; what they hate is looking up tables and itemizing | Keep the hardware scale; absorb the calculation labor |
| Supply-demand analysis | Missing the combined "whole meal at once → trustworthy total weight + automatic macros" | Defines the product-market fit |
| To be validated | Mixed Chinese-dish recognition, allocation accuracy, beginner trust, weekly-report priority | Prototype test-question checklist |
8 What to Validate Next
- Do users understand the "place the whole meal → get results" main path?
- Will the food-confirmation step become new friction?
- Results-page information hierarchy: total calories vs. macros vs. per-item grams?
- Is the auto-retry noticeable when recognition is unstable? Are the manual prompts for extreme cases (too dense / steam) clear?
9 Success Criteria
- Anyone can restate it within 30 seconds: the scale locks the weight · the AI splits the nutrition · the app builds the habit
- The main task (log a meal) has a clear main path + error recovery
- All six features covered, without adding steps just to "look sophisticated"
- Consistent accuracy narrative: every number traces back to the actual weighed weight W