🚀 My First Step Towards AI/ML Model Integration | Inspire Sphere

💡 The Spark

The idea that an ML model delivers exactly what you train it for was the spark I needed to take my first step into AI/ML.

While working on a web development project, I explored how AI algorithms could be integrated to generate meaningful predictions. That experience became the perfect opportunity for me to:

• Dive deeper into AI/ML,
• Polish my understanding,
• And start building real-world AI features.

---

🌐 The Project: Inspire Sphere

The project I worked on was Inspire Sphere — my first full-stack web app built with HTML5 + Node.js.

✨ It’s a quote generator + literature writing platform with user profiles.

---

🎯 The Idea Behind It

I wanted to predict the categories of quotes uploaded by users.

For this, I chose the Scikit-learn library and its submodules:

• ✅ sklearn.feature_extraction.text — to process the text content of quotes.
• ✅ TfidfVectorizer — to generate a sparse matrix of words with TF-IDF scores (importance based on frequency & repetition).

---

🧠 The Algorithm Used: Naive Bayes

I used Naive Bayes (MultinomialNB), which works well with word counts.

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.pipeline import make_pipeline
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report, accuracy_score
import joblib


from Dataset.load import quotes, labels

X_train, X_test, y_train, y_test = train_test_split(quotes, labels, test_size = 0.2 , random_state=42)

• 📊 The dataset (quotes + categories) was sourced from Kaggle.
• 🔎 The classifier learned to assign categories to user-generated quotes.

---

🗂️ Clean Code Structure

Even though Python is concise, I maintained a modular directory structure for clarity:

• Separate submodules for preprocessing, training, and model storage.
• This keeps the AI/ML branch of Inspire Sphere scalable and maintainable.

---

🔗 Integration with Node.js Using FastAPI

Once trained, the model was saved as a .pkl file using joblib.

For integration:

• 🖥️ I built a FastAPI POST API in Python.

app = FastAPI()

model = joblib.load(model_path)

class Quote(BaseModel):
    text:str

@app.post("/predict")
def predict(quote:Quote):
    prediction = model.predict([quote.text])
    return {"category":prediction[0]}

• 🔌 Node.js called this API to fetch predictions.

 async getCategory(quoteText) {
        try {
        const response = await axios.post(`${process.env.ML_SERVER_URL}/predict`, {text:quoteText});
        const category = response.data.category;
        return category;
        }
        catch (error) {
            console.error("Error While Getting Category", error);
            return null;
        }
    }

• 🌍 The API was hosted as a web service using Render.

This way, my Node.js app could consume AI predictions seamlessly.

---

✅ The Outcome

The ML model was able to:

• Predict categories of quotes uploaded by users.
• Enrich the HTML pages of Inspire Sphere with intelligent categorization.

---

🔧 This Is Just the Start

This was just my first step into ML/AI model integration.

⚠️ The model is currently low on accuracy and needs improvement.

But it opened the door for me to explore more advanced:

• Models,
• Feature engineering,
• And AI-powered integrations.

Stay tuned — I’ll be sharing more updates as I refine Inspire Sphere’s AI/ML domain. 🚀

---

🤝 Over to You

Have you ever integrated an AI/ML model into a web app?

What stack did you use — and what challenges did you face?

Let’s talk in the comments 👇