Australian, New Zealand (Māori), Chinese, Indian, Italian, Greek, Thai, Vietnamese, Japanese, Korean, Lebanese, Mexican, French, Spanish, Indonesian
Dietary restrictions (halal, kosher, vegan, gluten-free, etc.)
Nutrition-optimized recipes

📅 Meal Planning

Weekly meal plans
Budget-aware
Calorie/macro targeting
Leftover management

💪 Fitness Integration

Apple Health & Google Fit sync
Workout logging
Calorie tracking

🍔 Restaurant Tracking

AU/NZ chain nutrition
Healthier alternatives

📦 Barcode Nutrition

Product lookup
Allergen checking

👶 Kids Nutrition

School lunch planning
Allergen-free meals

Usage

Inference API

python

import requests

API_URL = "https://api-inference.huggingface.co/models/vasu24/groot-inosh-v3"
headers = {"Authorization": f"Bearer {YOUR_HF_TOKEN}"}

def query(payload):
    response = requests.post(API_URL, headers=headers, json=payload)
    return response.json()

# Example: Add pantry item
output = query({"inputs": "Add 500g chicken breast to pantry"})
print(output)
# Expected: {"action": "add_pantry", "item": {"name": "Chicken breast", "quantity": 500, "unit": "g", ...}}

# Example: Recipe suggestions
output = query({"inputs": "Show me quick Italian recipes under 30 minutes"})
print(output)

Load Locally

python

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained("unsloth/Llama-3.2-1B-Instruct")
tokenizer = AutoTokenizer.from_pretrained("unsloth/Llama-3.2-1B-Instruct")

# Load adapter
model = PeftModel.from_pretrained(base_model, "vasu24/groot-inosh-v3")
model = model.merge_and_unload()

# Generate
messages = [
    {"role": "system", "content": "You are GROOT, a kitchen AI assistant..."},
    {"role": "user", "content": "Add 500g chicken to pantry"}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=500)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Training Details

Dataset: 29,000 samples
Categories: 9 feature types
Iterations: 1000 (best checkpoint: 300)
Training Loss: 0.177
Validation Loss: 0.176
Method: LoRA fine-tuning on Llama 3.2 1B

JSON Response Format

GROOT responds with structured JSON for actions:

Pantry:

{"action": "add_pantry", "item": {"name": "Chicken breast", "quantity": 500, "unit": "g", "category": "Meat", "location": "fridge"}}

Recipes:

{"action": "suggest_recipes", "recipes": [{"name": "Pasta Carbonara", "time_minutes": 25, "difficulty": "easy", "ingredients": [...]}]}

Shopping:

{"action": "create_list", "store": "Woolworths", "items": [{"name": "Milk", "quantity": 2, "unit": "L"}]}

Limitations

Trained specifically for AU/NZ market
Uses metric units only
Prices in AUD/NZD
May not be accurate for other regions

License

Apache 2.0

Contact

Developer: Vasu Kapoor
App: iNosh - Smart Pantry Management
GitHub: Coming soon

Citation

bibtex

@misc{inosh-ai-2025,
  title={iNosh AI: Kitchen Assistant for Australia & New Zealand},
  author={Vasu Kapoor},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/vasu24/inosh-ai}
}