SmartPlant®

GardenSage™ — Your Personal Horticultural Expert

Mobile App
iOS
Android
AI Product
API Development

Visit Live Site

Scroll to read

Results that moved the needle.

01 -1
Native platforms shipped at parity

02 -45
Architecture nodes orchestrating the agent

03 -1M+
Plants identified across the platform

Challenge

Home gardeners had to guess. Plant identification, care schedules, pest diagnosis, and product recommendations lived in separate apps, blogs, and YouTube videos. SmartPlant needed one mobile experience that combined visual plant ID with conversational expertise, location-aware care, and an integrated product finder — across both iOS and Android, with parity.

Approach & Solution

We designed and built GardenSage™, the AI-powered companion inside the SmartPlant app. Snap a photo to identify, ask a free-form question to get an answer from a horticultural agent, and get a personalized care calendar tied to your hardiness zone. The product also surfaces relevant products inline — turning every care question into a path to purchase without breaking the conversation. Under the hood: a 5-step retrieval pipeline orchestrating 2 LLMs (GPT-4o-mini + GPT-4.1), 8 data sources (plant/care/products/zones), a clarification gate, and voice-mode TTS via ElevenLabs — all glued together by a custom SPH server with SSE streaming.

A 46-node pipeline, live.

The actual production architecture behind GardenSage™ — built by Ash. 5-step pipeline, 2 LLMs, 8 data sources, voice + photo + text modes, all orchestrated through our SPH server with SSE streaming. Hover any node on the live diagram to explore.

GardenSage Architecture — Traffic Flow

👤
User

BROWSER / MOBILE

💬
GardenSage Text

REACT + VITE

🎤
GardenSage Live

VOICE MODE

📷
Image Capture

IDENTIFY / SUGGEST


SSE Endpoint

SPH SERVER

🔄
Orchestrate v1

SPH SERVER

🔊
TTS Proxy

SPH SERVER

🔑
Voice Token

SPH SERVER

📋
Session CRUD

SPH SERVER

💡
Generate Tips

SPH SERVER

🎯
Intent Classifier

SPH SERVER

🧠

  1. Planner

INTENT PARSE + QUERY BUILD

🔬
Intent Parser

NLU EXTRACTION

📝
Query Generator

STRUCTURED PARAMS

💭
Context Analyzer

CONVERSATION HISTORY

🔍
2. Fetcher

PARALLEL DATA RETRIEVAL

🌱
Plant Database

SPECIES + VARIETIES

🧬
Genus Index

TAXONOMY + CLASSIFICATION

💧
Plant Care DB

GROWING GUIDES

🏪
Retailers DB

NURSERIES + INVENTORY

🛒
Products DB

FERTILIZERS + TOOLS

🌿
Plant Manager API

SPH SERVER

🗺️
USDA Zone Index

HARDINESS DATA

🖼️
Image Repository

CLOUDFRONT CDN

⚠️
Ambiguity Gate

CATEGORY VALIDATOR

📊
3. Ranker

MULTI-SIGNAL SCORING

📷
Image Quality

+4 POINTS

🎯
Name Match

+10 POINTS

🌡️
Zone Compat

+3 POINTS


Attribute Fit

+2 EACH


Brand / Source

+1 POINT

✍️
4. Synthesizer

GPT-4.1 RESPONSE GEN

🔁
Tool Use Loop

UP TO 4 ITERATIONS

📐
Response Shapes

5 VARIANTS

🖼️
5. Enrich

PHOTOS + MARKERS

🤖
GPT-4o-mini

OPENAI

🧬
GPT-4.1

OPENAI

📸
Wikimedia Commons

PHOTOS API

🗣️
ElevenLabs

TTS + CONVAI AGENT

🔐
Firebase Auth

AUTHENTICATION

🌤️
OpenWeatherMap

GEOCODE + WEATHER

📈
Mixpanel

ANALYTICS

☁️
CloudFront CDN

PLANT IMAGES

🗄️
MongoDB Atlas

SESSIONS + MESSAGES

💾
localStorage

CLIENT CACHE


Clarification UI

USER DISAMBIGUATION

46 nodes
Total Nodes

5 steps
Pipeline

2 LLMs
AI Models

8 DBs
Data Sources

2 gates
Clarification Loops

3 modes
Text / Voice / Photo

12+ functions
SPH Server Functions

Inside the product

A look at the live experience.

Tech Stack

Built with care.

Swift
Kotlin
React Native
Node.js
Claude API
GPT-4o-mini
GPT-4.1
ElevenLabs
MongoDB Atlas
Firebase Auth
Stripe

Next case study