Hero Screen
Mission Briefing
Archive Logs
Multi-modal capture and Analysis
Sun, Water, Seed and Soil feedback
Mission Progress

Plate Bestie

Your Farm-To-Table tool from Mars, powered by Gemini 3 Flash that turns real-world plant care into a survival mission.

Inspiration

Urban gardening often feels like a high-stakes guessing game. For city dwellers in small spaces, managing soil chemistry, inconsistent light, and watering schedules can be overwhelming. While many plant care apps exist, most are static databases that feel more like homework than help.

We asked: What if plant care felt like having a best friend who could actually see what you’re seeing?

Plate Bestie was born from the idea of Playful Productivity — combining game-inspired progress loops with a multimodal AI companion from Mars, Me-Lap, an intergalactic botanist learning Earth’s ecosystems alongside the user.

What It Does

Plate Bestie is a phygital (Physical + Digital) gardening companion that turns real-world plant care into interactive missions.

The core interface is a “Plate” divided into four zones:

Seed – Tracks growth stage and unlocks species-specific guidance
Soil – Evaluates substrate quality and estimated pH
Water – Guides hydration with real-time feedback
Sunlight – Assesses light exposure to find the “Goldilocks” zone

Instead of static tips, Me-Lap actively collaborates with the user:

Analyzes live camera input
Highlights areas of concern using visual grounding
Responds with native speech
Converts observations into structured care guidance
Rewards real-world actions with visible in-app progress

Players don’t just play — they adjust their actual plants at home based on what they learn.

How We Built It

Plate Bestie is built on Gemini 3 Flash, leveraging its native multimodality and low-latency reasoning to create an agentic feedback loop that feels alive and responsive.

Multimodal Input with Human-in-the-Loop (HITL) Staging

Users can stage multiple inputs before analysis:

Camera images of their plants
Optional voice notes describing concerns
Manual observations
Human review and feedback integration
Single-Source Ground Truth Protocol: To ensure maximum diagnostic accuracy, Plate Bestie utilizes a "Single-Source Ground Truth" protocol. When a file is uploaded, it becomes the primary diagnostic subject to prevent cross-contamination of visual data. Users can subsequently append environmental snapshots to provide habitat context, ensuring the AI focuses first on the specimen's specific cellular health.

These inputs are bundled into a single Gemini request, allowing the model to reason across vision, speech, and text simultaneously.

Structured Reasoning Output

Gemini returns a strict JSON response containing:

Solar exposure
Growth stage
Soil condition and estimated pH
Hydration status
A short mission summary from Me-Lap

This structured output directly powers the UI, game state, and reward logic.

Digital Plant Passport

Each analysis generates a persistent Plant Passport — a futuristic ID card showing Sun / Seed / Soil / Water status plus a mission log. Results are stored locally so users can track progress across sessions and return to prior analyses.

Visual Grounding

When Gemini detects issues (such as leaf discoloration or dry soil), bounding data is used to highlight regions directly on the camera feed, creating shared visual context between the AI and the user.

Agentic Game Loop

Gemini acts as a lightweight “game master.” Successful care actions trigger React state updates, achievements, and mascot state changes — reinforcing real-world behavior rather than replacing it.

Design Workflow

UX was designed in Figma with a custom component system, expressive mascot states, and rounded geometry, then implemented in Google AI Studio with React + Tailwind.

What We Learned

Grounded AI builds trust. Users engage more when the AI can visually point to what it’s discussing.
Low latency matters. Gemini 3 Flash made real-time interaction and mascot immersion possible.
Structured outputs beat free text. JSON-based responses dramatically simplified UI integration and persistence.
Gamification works best when it reinforces real behavior. Users were more likely to water or move plants when actions were tied to progress.
Data Persistence & Mission History. Chat interaction was not possible with mock archival logs because storage limitations were optimized for actual user scans for this project. We learned we needed to create multiple pathways to account for different data retrieval workflows.

What’s Next

Meteorological data API integration
Environmental sensing via IoT (light and moisture)
Long-term plant progression and seasonal events and causal analysis
Community garden sharing
Deeper mascot evolution based on user success
Retail API strategy

Tech Stack

Model: Gemini 3 Flash (Multimodal reasoning, Visual Grounding, Function Calling)
Frontend: React + Tailwind CSS
Design: Figma (Design System + Components)
Tools: Google AI Studio, Adobe Firefly