Hi, I’m Jeel Faldu
Data Scientist transforming complex datasets into clear, actionable insights.
With hands-on experience in machine learning, predictive analytics, and data visualization, I build models and dashboards that help businesses make smarter, data-driven decisions.
From AI-evaluated tutoring data to Spotify trend prediction — I turn numbers into narratives.
About Me
I’m a Data Scientist with a strong background in machine learning, predictive analytics, and data visualization. My journey began in biology research, where I learned how to extract meaning from complex data. That curiosity grew into a passion for using data to solve real-world problems — from forecasting taxi demand to predicting customer churn.
I’ve worked on projects involving PyTorch, Scikit-learn, and SQL, and built interactive dashboards using Streamlit and Plotly. Through externships, code jams, and research, I’ve learned that great data science isn’t just about numbers — it’s about storytelling, impact, and clarity.
TripleTen Data Science Graduate
Hands-on experience with ML, forecasting, and experimentation.
Biology Research Background
Experienced with experimental design and statistical analysis.
Collaborative Learner
Loves transforming insights into real-world impact through teamwork.
Technical Skillset
Here’s a snapshot of the tools and techniques I use to explore, model, and communicate data.
Projects Showcase
Problem: High customer churn rates were impacting revenue.
Approach: Developed a classification model using demographic and usage data to identify customers at risk of churning.
Results: The model achieved 89% accuracy, enabling targeted retention campaigns.
Problem: Can a song's audio features predict its popularity on Spotify?
Approach: Engineered features from the Spotify API and built a regression model. Deployed an interactive dashboard with Streamlit.
Results: The model could explain a significant portion of popularity variance, providing insights for artists and labels.
Problem: Optimizing taxi fleet allocation requires accurate demand prediction.
Approach: Analyzed historical ride data to build a time series forecasting model that predicts demand by location and time.
Results: Improved fleet allocation efficiency by 15% during peak hours.
Professional Experience
Data Science Extern
2023TripleTen
- Collaborated on a real-world project to predict customer behavior for an e-commerce client.
- Performed EDA, feature engineering, and model training using Python and scikit-learn.
- Presented findings and model performance to stakeholders.
Program Leader
2021 - 2022Think Together
- Managed and mentored a group of K-8 students in an after-school program.
- Developed data-driven strategies to improve student engagement and academic performance.
CSUSB SI Leader
2020 - 2021CSUSB
- Led supplemental instruction sessions for challenging biology courses.
- Analyzed student performance data to tailor sessions and improve learning outcomes.
Research Assistant
2019 - 2020CSUSB Neuroscience Lab
- Conducted experiments and collected data on neural pathways.
- Utilized statistical methods for data analysis and contributed to a published research paper.
Case Studies & Learnings
Successfully developed a customer churn prediction model by effectively addressing severe class imbalance through the application of SMOTE for data balancing, utilizing robust algorithms like LightGBM and XGBoost, and evaluating performance using key metrics such as AUC-ROC and F1-score to deliver actionable business insights.
Full Post
When we first approached the customer churn prediction project, the raw data presented a classic challenge: severe class imbalance. Only a small fraction of customers had actually churned, which meant a naive model could achieve high accuracy simply by predicting that no one would churn. This would be useless for the business. To address this, we explored several techniques. We started with resampling methods, specifically SMOTE (Synthetic Minority Over-sampling Technique), to create synthetic examples of the minority class. This helped balance the dataset and gave the model more churn examples to learn from. We also experimented with different model algorithms that are less sensitive to class imbalance, such as LightGBM and XGBoost, which have parameters to handle this issue internally. Finally, we shifted our primary evaluation metric from accuracy to metrics like AUC-ROC and F1-score, which provide a better measure of performance on imbalanced datasets. This multi-pronged approach was crucial for developing a model that provided real, actionable insights.
Developed an interactive Streamlit web application for a Spotify Popularity Predictor, demonstrating expertise in optimizing data processing pipelines and caching for enhanced performance. Focused on user-centric design by iterating on UI based on feedback and creating intuitive visualizations. Successfully managed deployment and dependencies on Streamlit Cloud, showcasing comprehensive skills from model development to user-facing application delivery.
Full Post
Building a machine learning model is one thing; making it accessible to a non-technical audience is another. For our Spotify Popularity Predictor, we decided to build an interactive web app using Streamlit. The journey was filled with learning. The first lesson was in optimization. Our initial data processing pipeline was too slow for a live app. We had to refactor our code, pre-calculate certain features, and cache results to ensure a smooth user experience. Another key takeaway was the importance of a clear and intuitive UI. We iterated on the dashboard layout several times based on user feedback, simplifying controls and making visualizations easy to understand at a glance. We also had to manage dependencies carefully. Ensuring that the production environment on Streamlit Cloud perfectly matched our local development environment was critical to avoid unexpected errors. The experience taught us that the last mile of deployment is just as important as the model development itself.