mealera

Mealera: Nigerian Food & Wellness Conversation Model

---

Mealera is a state-of-the-art conversational AI model designed to understand, classify, and respond to queries about Nigerian food, meal planning, health, shopping, and cultural dietary needs. Built on DistilBERT and fine-tuned on thousands of real-world, culturally rich conversations, Mealera empowers digital food platforms, health apps, and smart assistants to deliver context-aware, locally relevant, and health-conscious experiences for users in Nigeria and beyond.

---

🚀 Model Overview

• Model Name: mealera
• Author: divoishim
• Architecture: DistilBERT (Sequence Classification)
• Languages: English (Nigerian context, with local language and cultural adaptation)
• License: Apache 2.0
• Tags: Text Classification, Food, Health, Nigeria, Meal Planning, Vendor Recommendation, Allergy, Budget

---

🥗 What Can Mealera Do?

• Classify user queries into meal planning, health advice, shopping, vendor search, allergy, and more
• Understand Nigerian food culture, local language, and dietary restrictions
• Power chatbots, meal planners, health apps, and food e-commerce platforms

---

🧑‍💻 How to Use

Load the Model and Tokenizer

from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("divoishim/mealera")
tokenizer = AutoTokenizer.from_pretrained("divoishim/mealera")

Load the Label Encoder for Context Mapping

from huggingface_hub import hf_hub_download
import pickle
label_path = hf_hub_download(repo_id="divoishim/mealera", filename="label_encoder.pkl")
with open(label_path, "rb") as f:
    label_encoder = pickle.load(f)

Predict a Query's Context

import torch
query = "What can I cook for a family of four with 2000 naira?"
inputs = tokenizer(query, return_tensors="pt", truncation=True, padding=True)
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits
    predicted_class_id = logits.argmax().item()
print("Predicted context:", label_encoder.inverse_transform([predicted_class_id])[0])

---

🏷️ Context Labels (Intents)

The model predicts one of several context labels, including:

• recipe_recommendation: Suggesting recipes or meal ideas
• casual_chat: General conversation or greetings
• greeting: Salutations and opening messages
• vendor_recommendation: Finding food vendors or markets
• budget_meal: Affordable meal suggestions
• health_advice: Nutrition and health-related queries
• dietary_restrictions: Special diets (e.g., allergies, intolerances)
• shopping_list: Generating shopping lists
• meal_plan: Weekly/daily meal planning
• allergy_concern: Allergy and food intolerance questions

(See label_encoder.pkl for the full list and mapping.)

---

📦 Training & Evaluation

• Base Model: distilbert-base-uncased
• Training Data: 2,248+ annotated Nigerian food and wellness conversations
• Contexts: 10+ unique, culturally relevant categories
• Metrics:
  • Accuracy: 87.1%
  • F1 (macro): 0.84
  • F1 (weighted): 0.87
• Hardware: Google Colab GPU (T4)
• Training Time: ~6.5 hours (3 epochs)

---

✅ Intended Use Cases

• Food and meal planning chatbots
• Health and nutrition assistants
• Vendor and market recommendation systems
• Smart shopping list generators
• Culturally aware food apps for Nigeria and West Africa

---

⚠️ Limitations & Risks

• May misclassify queries with ambiguous or overlapping intent
• Best for English queries with Nigerian context; limited for other languages
• Not a substitute for professional medical advice
• Model may reflect biases present in the training data

---

🌍 Model Sources & Links

• Model Repo
• Training Data (private)
• Contact for Collaboration

---

📄 Citation

If you use Mealera in your research or product, please cite:

@misc{mealera2025,
  title={Mealera: Nigerian Food & Wellness Conversation Model},
  author={Divine Oshim},
  year={2025},
  howpublished={\url{https://huggingface.co/divoishim/mealera}}
}

---

🤗 Model Card Authors

• Divine Nzeh (divoishim)

---

💬 Questions or Feedback?

Open an issue on the Hugging Face repo or email [email protected].

---

Mealera: Empowering food, health, and culture—one conversation at a time.