Cookonomics

Co-cook with AI

Inspiration

We were inspired by the daily challenges of one-person households juggling tight budgets, time constraints, and healthy eating. Cookonomics was created to solve the burden of complex meal planning with AI, helping users live healthier and more economical lives.

What it does

Cookonomics is an AI-powered service that processes a user's natural language prompt containing their personal profile, such as age, body indicators, budget, and food preferences. In response, the system generates a fully personalized meal plan with a date range and an optimized shopping list tailored to the user's budget and nutritional goals. Users can also provide feedback by uploading photos of their cooked meals, which the system uses to continuously refine future recommendations.

How we built it

We architected Cookonomics as a sophisticated multi-agent system where specialized AI agents collaborate to tackle different aspects of meal planning. The system follows a carefully orchestrated workflow that begins when a user provides their personal profile, weekly budget, and food preferences through natural language input.

Our Meal Planner agent serves as the initial orchestrator, processing user requirements and generating preliminary meal suggestions. This connects to our Shopping Agent, which queries BigQuery databases to gather real-time food pricing and nutritional information. The Budget Analyzer then evaluates whether the proposed meals fit within the user's financial constraints, providing feedback to refine the meal selection through an iterative optimization loop.

Once the budget-optimized meal plan is finalized, our Meal Summarizer creates comprehensive meal summaries with detailed cooking instructions. The Meal Extractor then generates visual representations of the planned meals using Google's Gemini AI, providing users with appetizing images alongside their meal plans and shopping lists.

A critical innovation in our system is the implementation of semantic search using vector embeddings on ingredient data. Rather than relying on simple keyword matching, this allows us to understand the nuanced relationships between different ingredients and accurately translate abstract user preferences like "I want something comforting but healthy" into concrete recipe recommendations.

Challenges we ran into

One of our most significant challenges was bridging the gap between abstract user preferences and concrete, actionable recipes. People often describe their food desires in subjective terms—"something light and fresh" or "comfort food that's not too heavy"—but recipes require specific ingredients and measurements. We solved this by implementing vector embeddings that capture the semantic relationships between ingredients, enabling our system to understand taste profiles beyond literal descriptions.

From a technical architecture standpoint, designing a robust multi-agent workflow presented considerable complexity. Each agent needed to perform its specialized function while maintaining loose coupling with other system components. We had to carefully minimize dependencies between agents to ensure the system remained resilient against individual component failures and could gracefully handle fallback scenarios when needed.

Accomplishments that we're proud of

Our proudest accomplishment is the successful implementation of a semantic search system using vector embeddings for ingredients. This allows us to accurately translate user preferences into relevant recipes by searching based on taste profiles, not just keywords. We also successfully architected a robust, decoupled multi-agent workflow, which ensures the system remains operational and resilient even if an individual agent encounters an issue.

What we learned

The most valuable lesson from building Cookonomics was understanding the critical importance of workflow design in multi-agent systems. While individual agent capabilities are important, the overall system's effectiveness depends heavily on how these agents communicate and coordinate with each other. We learned that a well-architected workflow serves as the foundation for system stability and scalability.

What's next for Cookonomics

We aim to expand our solution to reach a wider audience and foster a vibrant, health-focused community where users can share their lifestyles and creative recipes. Looking ahead, we also plan to support individuals with specific health conditions by helping them enjoy nutritious meals tailored to their needs—potentially in collaboration with healthcare professionals for medically-informed guidance.