AI Grocery and Meal Planning Manager

Product challenge

Households do not experience pantry tracking, deciding what to cook, recipes, and grocery shopping as separate problems. The product opportunity was to connect them into one useful loop, while keeping AI grounded in live household context rather than generating generic meal suggestions.

Product and engineering approach

Build around household context

• Built a visual inventory system for organizing fridge, pantry, and staple items across named shelves.
• Added AI-assisted stocking that converts a natural-language description of groceries into structured inventory data.
• Supported multiple households with isolated inventory, recipes, meal plans, and preferences.

Give AI a clear job in the workflow

• Created a weekly planner that recommends meals based on available ingredients, preferred dishes, dietary preferences, and variety.
• Generated structured, easy-to-follow recipes and highlighted the ingredients still needed for each meal.
• Added voice-driven recipe creation and persistent recipe-image generation with progress that survives refreshes.
• Kept model providers behind the product boundary so Gemini, OpenAI, or Anthropic could be used without rewriting the household experience.

Close the loop to action

• Automatically consolidated missing recipe ingredients into a practical grocery list.
• Integrated Swiggy Instamart into the replenishment flow so users can move from meal planning to grocery ordering with less manual searching.
• Added authenticated persistence with PostgreSQL and Prisma so household state remains available across sessions.

Outcome

• Connected pantry awareness, recipe discovery, meal planning, and grocery ordering in one workflow.
• Reduced repeated planning by deriving meals and shopping needs from household inventory and preferences.
• Demonstrated an applied-AI pattern I value: combine probabilistic generation with structured data, persistent state, user control, and a concrete action outside the model.