Growcery uses AI and computer vision to analyze images of crops and produce.
With a single photo, users can instantly learn about freshness, quality, or possible diseases, along with treatment advice or storage recommendations — all accelerated by AMD ROCm GPUs.

• For shoppers: Detects freshness, spoilage, and optimal storage conditions for fruits and vegetables.
• For farmers: Identifies plant diseases and gives actionable treatment or prevention guidance.
• For everyone: Delivers natural, human-readable insights generated by Gemini 2.5 Flash, powered by the Google GenAI SDK.

[ Next.js (TypeScript) Frontend ]
        │
        ▼
[ FastAPI Backend (Python + PyTorch on ROCm) ]
        │
        ▼
[ Gemini 2.5 Flash (GenAI API) ]

1. Image Capture / Upload — Users upload a photo from the web app.
2. Inference Stage — FastAPI sends the image to PyTorch models (MobileNetV3, EfficientNetV2) accelerated by AMD ROCm.
3. Reasoning Stage — The backend calls Gemini 2.5 Flash through the Google GenAI SDK for TypeScript to interpret raw predictions.
4. Response Delivery — Gemini returns descriptive advice (e.g. “This tomato shows early blight. Treat with copper fungicide and avoid overhead watering.”)
5. Frontend Display — Results and latency stats are displayed in the web UI.

• Next.js / React (TypeScript)
• TailwindCSS
• Image upload + result visualization + latency display

• FastAPI (Python 3.10+)
• PyTorch with ROCm 7.0
• Google GenAI SDK (TypeScript) for Gemini calls
• JSON I/O pipeline connecting model inference and reasoning

• Backend: Deployed on AMD ROCm 7.0 cloud instance
• Frontend: Deployed on Vercel (Next.js)
• Inference: ONNX-compatible PyTorch pipeline
• Reasoning: Gemini 2.5 Flash API calls via TypeScript

Growcery was developed during the KnightHacks Hackathon by:

• Alexander Nardi
• Natalia Cano
• Tai Williams
• Ray Arcand

• 🌐 Devpost Submission
• 🧠 Powered by Google Gemini 2.5 Flash & AMD ROCm 7.0