Ecocart is a full-stack sustainability platform that helps users evaluate the environmental and ethical impact of food and household products. It combines structured product data, semantic search, and an intelligent sustainability agent to make responsible purchasing decisions easier and faster.

Ecocart allows users to:

• Search for food and household products using semantic similarity
• View sustainability scores with clear reasoning
• Get concrete alternative suggestions
• Create and manage persistent shopping lists
• See per-item sustainability assessments inside saved lists

The system is built for speed, clarity, and real-world usability. Sustainability insights are integrated directly into search results and shopping workflows rather than presented as isolated reports.

Ecocart is built with Next.js as a full-stack framework, providing:

• File-based routing
• Server-side rendering
• Unified frontend and backend logic

Product data originates from Open Food Facts. The pipeline includes:

• Preprocessing scripts to extract essential attributes and reduce dataset size
• Cleaned data stored in Supabase with a relational schema
• Supabase also handles authentication and persistent shopping lists

Search is powered by embeddings and a vector database using Pinecone:

• Separate embeddings for product name, brand, and categories
• Weighted multi-signal similarity ranking with brand-first prioritization
• Deduplication of identical unbranded entries

Flow: User query → vector search retrieves best-matching product IDs → relational DB loads full product records → sustainability agent can evaluate selected products.

The sustainability agent is a tool-using LLM system built with OpenAI (e.g. gpt-5-mini). It evaluates products based on:

• Environmental impact
• Ingredients and sourcing
• Ethical considerations
• Brand-level sustainability context

Input: Product name, brand, categories, labels, ingredients, ecoscore (if available).

Output (structured JSON): Verdict (good / moderate / poor), score (0–100), reasoning, better alternatives, and summary tags.

The agent runs in a controlled multi-turn loop and can call tools such as product detail retrieval (Open Food Facts) and web search for sustainability context. Outputs are validated before being returned to the frontend.

Users can create, rename, and delete lists, and add or remove products. When a product is added, the sustainability agent runs in the background and stores the score and tags with the item, so users can evaluate their entire cart at a glance.

• Node.js (v18+)
• Supabase project
• Pinecone index
• OpenAI API key

Copy .env.example to .env.local and fill in:

# Supabase (https://supabase.com/dashboard/project/_/settings/api)
NEXT_PUBLIC_SUPABASE_URL=https://your-project-id.supabase.co
SUPABASE_ANON_KEY=your-anon-key
# For seeding/sync scripts
SUPABASE_SERVICE_ROLE_KEY=your-service-role-key

# Search + sustainability
OPENAI_API_KEY=your-openai-api-key
PINECONE_API_KEY=your-pinecone-api-key
# Optional: PINECONE_INDEX_NAME=hopperhacks, MASTER_TABLE=api_items

# Optional: for agent web search (sustainability context)
SERPER_API_KEY=your-serper-api-key

npm install
npm run dev

Open http://localhost:3000.

Note (Windows): The dev script uses next dev --webpack to avoid a known Turbopack + PostCSS timeout when processing app/globals.css. To use Turbopack, run npx next dev instead.

• Seed products: npm run db:seed-api-items or npm run db:seed-products (requires CSV and Supabase env)
• Sync vectors to Pinecone: npm run db:sync-vectors (requires OpenAI, Pinecone, Supabase env)

• Performance first: Fast search and responsive scoring
• Structured outputs: Schema validation for reliability
• Explainability: Clear reasoning behind every score
• User experience: Clean results, deduplication, actionable alternatives
• Real-world impact: Sustainability integrated into actual shopping behavior

• Large-scale data preprocessing
• Embedding generation and vector storage (Pinecone)
• Multi-namespace vector search with weighted ranking
• Tool-augmented LLM orchestration
• Structured AI output validation
• Full-stack system design with Next.js

• Expanded sustainability datasets
• Personalized scoring preferences
• Hybrid keyword + embedding retrieval
• Browser extension and barcode scanning
• Deeper brand-level sustainability analysis

Ecocart is designed to turn sustainability from abstract information into an actionable part of everyday purchasing decisions.