Verdura AI

Verdura AI connects local farmer market produce with personal wellness goals using real-time data and AI—helping users eat smarter, live healthier, and support local agriculture.

Comment

Verdura AI – Eat Well. Live Fresh. Powered by AI.

Inspiration

Verdura AI was inspired by the desire to create a bridge between local, seasonal food availability and personalized wellness recommendations. While nutrition apps often focus on tracking macros or counting calories, they rarely integrate with real-world data like farmers markets, seasonal produce, or community-based food systems. This project reimagines food technology through the lens of location, seasonality, and AI, helping users align their eating habits with both their wellness goals and what’s fresh locally.

What it does

Verdura AI is an AI-powered farm-to-wellness platform that allows users to:

  • Search for local produce by ZIP code and discover what’s fresh near them
  • Ask AI-driven wellness questions such as “What should I eat for better sleep?”
  • Generate smart seasonal food bundles based on available produce and wellness goals
  • Receive a personalized 7-day meal plan using seasonal ingredients
  • View a demo marketplace of local farmers and vendors with product listings
  • Analyze nutrition from either typed ingredients or food images using AI

It combines data science, AI, and intuitive UI to deliver a complete food and wellness experience.

How we built it

The project is composed of a Next.js frontend and a FastAPI backend, connected through REST endpoints and powered by Google’s Gemini AI models.

Frontend:

  • Next.js 15.2.4 with React 19 (App Router)
  • TypeScript for static type safety
  • TailwindCSS for styling
  • Radix UI and ShadCN for accessible UI components
  • Framer Motion for animations and transitions
  • ESLint and Prettier for code quality
  • Utility libraries: clsx, lodash, etc.

Backend:

  • FastAPI (Python 3.12) for REST API routes
  • Uvicorn as the ASGI server
  • Google Generative AI (Gemini 1.5 Pro & Flash) for AI-powered endpoints
  • Pydantic for request validation and typing
  • Pandas for data transformation and CSV parsing
  • Pillow (PIL) for image processing (Vision AI input)
  • Python-dotenv for secure environment management

Architecture:

  • TypeScript-based frontend using Next.js with modular routing
  • Python backend exposing LLM and vision-powered endpoints
  • JSON-based API communication between client and server
  • Local USDA dataset integration (CSV) for produce/farmers data
  • Designed for deployment via Vercel (frontend) and optionally Render/Railway for backend

Challenges we ran into

  • Real-time API limitations: USDA APIs lacked reliable or free real-time access. As a workaround, we sourced and cleaned static datasets from USDA.gov and built a parser to convert it into usable frontend JSON.
  • Data inconsistencies: Farmers market data varied in structure and completeness, requiring extensive data cleaning and schema mapping.
  • Prompt engineering: Formatting prompts for Gemini to generate structured, goal-relevant outputs required careful tuning and error handling.
  • Cross-stack debugging: Maintaining API contract consistency across TypeScript (frontend) and Python (backend) required close coordination.

Accomplishments that we're proud of

  • Built a working, AI-powered health assistant with real produce data in under 24 hours
  • Integrated both text and image-based macro estimators using Google’s Gemini APIs
  • Created a modular and extensible app architecture with seamless frontend/backend communication
  • Developed a clean, mobile-responsive UI with animated transitions and real-time UX
  • Leveraged public datasets to build a socially conscious and practical AI application

What we learned

  • How to build full-stack applications with modern web tools (Next.js + FastAPI)
  • Advanced prompt design for health-related reasoning using generative models
  • Strategies for transforming raw open data (CSV) into usable digital content
  • Image processing and Vision AI input handling with Gemini
  • The importance of building developer-friendly tooling (ESLint, TypeScript, modular design) for velocity and reliability

What's next for Verdura AI

  • Connect to the live USDA Farmers Market API for real-time produce and event data
  • Enable user authentication and saving of bundles, wellness goals, and plans
  • Add user-specific dashboards for tracking nutrition and seasonal eating
  • Allow farmers to register and manage their storefronts directly on the platform
  • Integrate macro and calorie tracking history using Firebase or another cloud backend
  • Finalize deployment for public access and expand beyond Chicago to nationwide coverage

Built With

Languages & Frameworks

  • TypeScript / React 19
  • Next.js 15.2.4
  • TailwindCSS, ShadCN, Radix UI
  • Framer Motion (animations)
  • FastAPI / Python 3.12
  • Pandas, Pillow, Pydantic

AI & APIs

  • Google Gemini 1.5 Pro (LLM) – wellness chatbot, bundles, macro text estimator
  • Gemini 1.5 Flash (Vision) – food image analysis for nutrition estimation
  • USDA.gov – Chicago farmers market CSV dataset
  • DiceBear API – avatar generation for farmer profiles

Dev & Tools

  • Uvicorn – ASGI server
  • dotenv – Environment variable loading
  • ESLint, Prettier – Code formatting and linting
  • Cursor IDE – AI-powered development environment
  • Postman / HTTP clients – API testing

Verdura AI was designed, engineered, and deployed by a solo developer in under 24 hours for a hackathon challenge, with a vision to make food intelligence accessible, seasonal, and local.

Built With

Share this project:

Updates