BudgetBite

Inspiration

As students, we kept running into the same problem every week: eating healthier felt expensive, and staying on budget meant repetitive or low-quality meals. We wanted to build something that turns meal planning into a practical, cost-aware workflow — not just a recipe browser. The core idea was to combine nutrition guidance with real supermarket pricing so users can see what they can actually afford before they ever set foot in a store.

What it does

Untitled is a budget-first meal planning app that generates a multi-day breakfast, lunch, and dinner plan based on your preferences. A 3-step wizard captures your budget, dietary restrictions, cuisine preferences, difficulty, and cooking time — then produces:

A full day-by-day meal calendar
Estimated cost per meal and weekly total
A consolidated shopping list with quantities and prices
Meal detail modals with ingredients, macros, and instructions
Regeneration controls to quickly replan using notes and include/avoid ingredient lists

How we built it

We built the app with Next.js 16 App Router, TypeScript, and Tailwind CSS.

Frontend: A landing page feeds into the wizard, which encodes preferences into a URL. The results page generates and renders the plan client-side via an API workflow.

AI workflow: An ingredient selection agent picks practical ingredients from the dataset, while a weekly planner agent generates meals in strict JSON format. The server validates responses with Zod schemas, enriches meals with pricing and images, and computes totals.

Cost and planning logic: Recipes are filtered by hard constraints (diet, difficulty), then scored by weighted factors — affordability, nutrition, cuisine fit, and ease — and assigned greedily into meal slots with variety checks. Shopping quantities are aggregated and priced against the cheapest matching products.

The scoring formula is: S = 0.40·A + 0.25·N + 0.20·C + 0.15·E, where A is affordability, N is nutrition match, C is cuisine match, and E is ease/time fit.

Challenges we ran into

AI JSON reliability: Models occasionally returned extra text around JSON, requiring hardened parsing and fallback extraction logic.
Budget realism: Per-serving costs can look affordable while real checkout totals are higher due to pack sizes — we had to calculate both theoretical consumption and actual shopping costs.
Ingredient normalization: Raw ingredient records had inconsistent naming and price fields, requiring careful deduplication and normalization.
Constraint balancing: Keeping plans varied, nutritionally sound, and within budget simultaneously required multiple rounds of iteration and retry logic.
UX during generation: Without a strong loading state, generation felt abrupt. We added staged progress indicators and animated skeletons to improve perceived responsiveness.

Accomplishments that we're proud of

Built an end-to-end pipeline that goes from user preferences to a fully actionable shopping plan
Integrated AI generation with strict schema validation and robust error handling
Implemented dual cost views and shopping aggregation logic that reflects real-world spending
Delivered a polished results experience with interactive meal detail modals and smooth loading animations
Kept the entire flow simple and fast — no sign-in required

What we learned

Reliable AI products need guardrails, not just prompts. Schema validation and retries are non-negotiable.
Cost modeling is nuanced. Users need clear distinction between per-serving estimates and real checkout totals.
UX details build trust. Loading states and control placement significantly affect how credible the app feels.
Strong type safety pays off. Shared data contracts made it much easier to evolve a multi-step workflow quickly.

What's next for Untitled

Live supermarket price sync and historical price tracking
Smarter budget optimization using pack-aware selection instead of simple heuristics
Personalized reranking based on meal feedback (likes, dislikes, swaps)
Allergy-safe and pantry-aware planning modes
Export options — shopping checklists, calendar views, and printable meal prep cards
Test coverage for scoring, costing, and schema validation paths