Smart Call Center Analyzer

A Portfolio Project by Sabeen Zehra

AI-powered insights using XGBoost, DistilBERT, and T5-Small

Overview Page

Dashboard with metrics (1.26M tweets, 10.9% churn, 0.48 avg. priority) and visualizations of sentiment and intent.

Overview Page Top Screenshot
Overview Page Bottom Screenshot

Integrated Analysis Page

Input: "this is awful, I want to cancel my account now!"
Output: High Churn NEGATIVE Cancellation High Priority

Integrated Analysis Page Layout
Integrated Analysis Output 1

Output: High Churn, NEGATIVE, Cancellation, High Priority

Churn Prediction Page

Input: "this is awful, I want to cancel my account now!"

Churn Prediction Page Layout
Churn Prediction Page Layout
Churn Prediction Output 1

Output: High Churn Risk (>50%)

Churn Prediction Output 2

Output: No Churn Risk (e.g., neutral tweet)

Sentiment Analysis Page

Inputs: "this is awful" and "this is great"

Sentiment Analysis Page Layout
Sentiment Analysis Output 1

Input: "this is awful" (NEGATIVE)

Sentiment Analysis Output 2

Input: "this is great" (POSITIVE)

Intent & Priority Page

Inputs: - "this is awful, I want to cancel my account now!" - "this is awful" - "this is great"

Intent & Priority Page Layout
Intent & Priority Output 1

Input: "this is awful, I want to cancel..." (Cancellation, High)

Intent & Priority Output 2

Input: "this is awful" (Complaint, Medium/High)

Intent & Priority Output 3

Input: "this is great" (Praise, Low)

End of Screenshots - Explore my journey in the About Me tab! 🚀

Created by Sabeen Zehra | View project on GitHub

About Me & My Project

The journey behind the Smart Call Center Analyzer

Hi, I'm Sabeen Zehra!

Currently in my final year of BS Data Science, somewhere between debugging models and fighting off imposter syndrome 😣.

I built this project as a personal milestone. No deadlines. No assignment prompts. Just me, my laptop, and an unhealthy number of browser tabs open at all times ☕💤.

The Smart Call Center Analyzer started as a portfolio idea but turned into a deep dive into real-world AI. From predicting churn to understanding emotions to prioritizing queries with GenAI, I aimed to make each module feel like it belongs in a real product.

I didn’t want fancy, I wanted sensible. And maybe, just maybe, something a recruiter might scroll through and think, “Okay... she gets it.” 🤞

🛠️ Building the Solution (The Fun Part!)
1

🧹 Data Detective Work

Cleaned 1.26M tweets, dealt with emojis, URLs, and noise. Real data is never textbook-clean!

Pandas, RegEx, patience
2

🔧 Feature Engineering Magic

Created 5,010 features with TF-IDF. Learned 'cancel' and 'awful' are key predictors!

Scikit-learn, TF-IDF, VADER
3

🤖 Model Training Marathon

Trained XGBoost for days, hit 97.3% F1-score. Jumped out of my chair when I saw it!

XGBoost, hyperparameter tuning
4

🧠 Adding Intelligence

Integrated DistilBERT and T5-Small. AI understanding emotions is mind-blowing!

Hugging Face Transformers
5

🌟 Bringing It to Life

Built this Streamlit app for interactivity. Great models need great UI!

Streamlit, Plotly, CSS
What I Learned (The Real Value)

🔬 Technical Skills

  • Python mastery: From scripts to ML pipelines
  • ML expertise: Feature engineering, model selection
  • NLP magic: Transformers and language understanding
  • Data visualization: Telling stories with numbers

💡 Life Lessons

  • Patience pays off: Good models take time
  • User experience matters: Accessibility is key
  • Documentation is key: For future me
  • Problem-solving mindset: Errors are opportunities
Achievements

97.3%

F1-Score for Churn

92%

DistilBERT Accuracy

1.26M

Tweets Processed

5,010

Features Engineered

📬 Connect with Me

🤝 Let’s Talk!

Thanks for stopping by. You’ve made this project real by reading this. If you’re into meaningful tech and a bit of chaotic creativity, let’s connect!

LinkedIn: sabeen-zehra-6635aa355

Email: syedasabeen583@gmail.com