Planta Livre

For Planta Livre we built Plant AI Agent — an intelligent procurement and budgeting system for the plant industry that turns unstructured requests into priced, exportable budgets.

The problem

Procurement requests arrive as emails, PDFs, spreadsheets, and photos, in Portuguese and English. Staff extracted plant names, quantities, and specifications by hand, then matched each line item against a catalogue of more than 10,560 plants — slow, error-prone, and impossible to scale during peak season.

Our approach

• Multilingual (PT/EN) request understanding with OpenAI GPT-4.1, extracting structured line items from unstructured emails, documents, and chat messages
• Semantic product matching with vector embeddings and cosine similarity across the 10,560+ plant database, robust to informal descriptions, synonyms, and naming variations
• Multi-format ingestion (emails, PDFs, Excel, DOCX, images) unifying every procurement channel into one pipeline
• A modern web interface (Next.js 15, React 19, TypeScript, FastAPI) with real-time budget tables, drag-and-drop interactions, and confidence-based indicators

Outcome

Plant AI Agent processes requests of 150+ line items with sub-second product matching across the full catalogue — significantly reducing budget creation time and cutting errors in procurement quotes.