Savr.

The problem

Existing meal planners (Mealime, PlateJoy, Plan to Eat) all work the same way: pick recipes → generate shopping list → assume you'll buy the ingredients. That model wastes money, wastes food, and ignores what's already sitting in the user's pantry — often the items closest to expiring.

The downstream pain is concrete: recipes don't match the dietary need; user preferences (maximize savings vs maximize variety) aren't honoured; budget gets blown when the system can't recover from a single expensive recipe. Users end up either over-spending or over-buying — and the food waste compounds week over week.

The solution

Savr inverts the model: pantry + budget first, recipes second. The user fills a structured form with weekly budget, number of meals, dietary needs, 2–3 cuisine preferences, existing pantry items, and a preference (maximize savings / balanced / maximize variety). The AI planning engine builds a plan that maximizes ingredient reuse, only adds new ingredients if still under budget, and produces a dynamic shopping list that excludes pantry items.

The AI tone is deliberately reliable meal-planning assistant, not chef influencer or chatty companion — supportive, efficient, clear, warm enough to make planning feel easy, but never performative.

How it works

Flow: structured input form (budget, meals/days, dietary needs, cuisines, pantry items, preference) → planning engine selects recipes that match dietary needs → reuses ingredients across meals per preference → estimates cost from approximate ingredient prices → checks against budget → if over budget, swaps recipes to reduce cost → outputs weekly plan + dynamic shopping list with reuse/waste metrics.

Primary model: GPT-4o-mini — chosen for Structured Outputs support (consistent JSON from template), speed, and affordability. The task profile fits this tier exactly: parse user constraints, select from recipe space, do structured generation, return JSON.

Automated evaluation: 12 schedule-coverage tests (under-fill, hybrid retry recovery, retry exhaustion, outside-window single retries, over-fill trim e.g. 19/15 pattern, exact-match no-trim) and 17 invariant tests running on every change to catch regressions.

Who it's for

Primary users: busy students, working professionals, and eco-conscious consumers who: (a) want to save money by planning better; (b) have specific dietary restrictions; (c) hate wasting food; (d) want planning to feel lightweight and guided, not like another spreadsheet.

Required inputs: weekly budget, dietary restrictions (1+), cuisine preferences (2–3), preference (reuse / balanced / variety). Optional: pantry items (item + quantity). All inputs are user-customizable so the planning engine adapts to changing budgets, dietary needs, or cuisine moods week over week.

Why it matters

AI-driven meal planning is growing at 28.10% CAGR (2025–2034) — driven by personalized health, dietary specificity, and the broader move toward conscious consumption. The incumbents (Mealime, PlateJoy) have proven the demand for AI meal planning but are still designed recipe-first, treating the shopping list as a side effect rather than the constraint.

Savr's strategic bet is constraint-driven planning: zero-waste, budget-first, time-aware — solving the *actual* meal-planning problem (food and money waste) rather than the surface problem (recipe variety). The friendly-beta launch approach mirrors that thesis — validate that the constraint-first framing resonates before scaling, instead of releasing-then-pivoting.

At a glance

Project

Savr

Built by

Mahima Pande

One-liner

Zero-waste, budget-first, time-aware AI meal planner — generates weekly plans that respect dietary needs, maximize pantry use, and stay under budget.