🏆 Start Here: Register & Get Aura Credits: Aura Agent Hackathon

[abimael.carrasquillo]

Jibarito Agent - Neo4j Aura Agents Hackathon Submission

Agent Name

Jibarito — Puerto Rican farmer/campesino
An agricultural intelligence agent for the US Caribbean & tropical regions

---

What It Does

Jibarito is a bilingual (Spanish/English) AI agent powered by Neo4j AuraDB that helps farmers, gardeners, agronomists, and researchers make data-driven agricultural decisions.

Aura Console Screenshots

---

Live Demo

YouTube Demo Video: https://youtu.be/22F9bWTdzvs

Features demonstrated:

• Real-time queries in English and Spanish
• Companion planting recommendations
• Production statistics & comparisons
• Dark mode UI with Jibarito branding
• Data attribution to USDA, CrophGraph & Neo4j

---

Live Application

Try Jibarito: https://jibarito.streamlit.app/

• Click example questions in the sidebar
• Type your own questions (English or Spanish)
• Get instant agricultural intelligence
• All responses powered by Neo4j AuraDB

Answers questions about:

• Crop Production Statistics — What's grown where, historical trends (USDA NASS data)
• Companion Planting — What plants grow well together with mechanisms & sources
• Pest & Disease Management — Organic solutions for agricultural pests
• Growing Information — Spacing, harvest times, climate requirements
• Regional Comparison — Compare production across 4 territories
• Bilingual — Ask in English or Spanish, get answers in your language

---

Dataset & Why a Graph Fits

The Data

• 92,218 Agricultural Production Records from USDA NASS Quick Stats
• 4 US Territories: Puerto Rico, US Virgin Islands, Hawaii, Florida
• 3 Census Years: 2018, 2022, 2023
• 228 Crops
• 262 Pests

Why Neo4j Graph is Perfect

Agricultural intelligence is fundamentally about relationships, not isolated data:

1. Companion Planting Networks → "What grows well WITH tomatoes?"
2. Crop Rotation Sequences → "What to plant AFTER tomatoes?"
3. Pest Management Relationships → "What pests attack tomatoes?"
4. Production Context → Commodity → AgRecord → Geography → County with full history
5. Semantic Discovery → Vector similarity on 92K production records

---

Final Graph Statistics

• 92K+ Nodes — territories, counties, commodities, plants, pests, operations
• 213K+ Relationships — production, companions, pests, rotations, geography
• Vector Index on 92,218 AgRecords (OpenAI embeddings, cosine similarity)

Graph Schema

Layer 1: Production Statistics

Source: USDA NASS Quick Stats

ClimateZone → Geography → County

AgRecord ──→ IS_COMMODITY  ──→ Commodity ──→ THRIVES_IN ──→ ClimateZone
         ──→ FOR_GEOGRAPHY ──→ Geography
         ──→ REPORTED_IN   ──→ County

CommodityGroup ──→ HAS_COMMODITY ──→ Commodity

OperationStats ──→ STATS_FOR ──→ Geography | County

Layer 2: Horticultural Intelligence

Source: CropGraph API + USDA PLANTS Database

Plant ──→ COMPANION_OF    ──→ Plant          (mechanism, strength)
      ──→ ANTAGONIST_OF   ──→ Plant         
      ──→ HAS_PEST         ──→ Pest           (severity, type)
      ──→ BELONGS_TO_FAMILY ─→ RotationFamily
      ──→ SAME_AS          ─→ Commodity      (bridge to Layer 1)

RotationFamily ──→ ROTATE_TO ──→ RotationFamily

---

Technology Stack

Component	Technology
Graph Database	Neo4j AuraDB
Agent Platform	Neo4j Aura Agent
Embeddings	OpenAI text-embedding-3-small (1536 dims)
Frontend	Streamlit 1.31.0
Deployment	Streamlit Cloud

---

Data Sources

Source	Type	License	Records
USDA NASS Quick Stats	Production statistics	CC0	92,218
USDA Census	Farm operations	CC0	Operations data
CropGraph API	Companion planting, pests, spacing	Open	91 crops
USDA PLANTS	Botanical traits	Public Domain	125 crops

---

Example Queries

English:

• "How many tomatoes were grown in Puerto Rico in 2022?"
• "What should I plant next to tomatoes?"
• "What pests attack peppers?"
• "What crops are grown in Hawaii?"

Spanish:

• "¿Cuántos tomates se produjeron en Puerto Rico en 2022?"
• "¿Qué debo sembrar junto a los tomates?"
• "¿Qué plagas atacan los plátanos?"
• "¿Qué cultivos se cultivan en Hawái?"

---

Architecture

User Query (Streamlit)
    ↓
OAuth2 Token Generation (Neo4j)
    ↓
REST API → Neo4j Aura Agent
    ↓
Agent Tools:
  • Similarity Search (vector embeddings)
  • Commodity Queries (Cypher templates)
  • Companion Plant Relationships
  • Pest Management Info
  • Growing Characteristics
    ↓
Neo4j AuraDB
  • 92K+ nodes
  • 213K+ relationships
  • Vector index (ag_embeddings)
    ↓
Clean Response → User

---

Key Decisions

Scope Selection

• 4 territories + 3 census years fitting AuraDB limits

Multi-Layer Design

• Layer 1 (Production Stats) — USDA NASS data for exact queries
• Layer 2 (Horticultural Intelligence) — CropGraph + USDA PLANTS for recommendations
• Separation enables both precise queries and discovery-based learning

Bilingual Support

• Puerto Rico + Spanish-speaking farmers require native Spanish support
• Aura Agent naturally handles language switching via system prompt

Vector Embeddings

• OpenAI text-embedding-3-small (1536 dims) chosen to match Aura Agent's default
• Enables semantic similarity search and agricultural records

---

How to Use Jibarito

1. Visit: https://jibarito.streamlit.app/
2. Choose: Click any example question in sidebar OR type your own
3. Wait: Agent processes your question using Neo4j knowledge graph
4. Learn: Get data-driven agricultural intelligence in your language

---

Links

• GitHub Repo: https://github.com/acarrasquillo/jibarito_agent

Built for the Neo4j Aura Agents Hackathon | June 2026