M.S. Applied Data Science student at USC | Graduate Research Assistant @ USC Information Sciences Institute (ISI) Working on deep learning, knowledge graphs, MCP integrations, and agentic AI pipelines. Previously an NSF REU researcher building a U-Net3+ CNN for tornado detection on radar data. Open to full-time entry-level roles in data science, ML engineering, and AI/software engineering.
Applied ML for real-world problems, from classical statistical learning through deep learning.
| Homework | Techniques | Dataset |
|---|---|---|
| KNN Classification | EDA, Classification, KNN | Vertebral Column |
| Regression | Linear/Multiple/KNN Regression, Hypothesis Testing | Combined Cycle Power Plant |
| Time Series Classification I | Time-domain Feature Extraction, Bootstrapping | AReM |
| Time Series Classification II | Binary/Multiclass Classification, Logistic Regression | AReM |
| Decision Trees | Cost-Complexity Pruning, Ridge, LASSO, Boosting | Acute Inflammations, Communities & Crime |
| Tree-Based Methods | Random Forest, SMOTE, XGBoost | APS Failure at Scania Trucks |
| Support Vector Machines | SVM, K-Means, Monte-Carlo Simulation | Anuran Calls (MFCCs) |
Final Project: Transfer learning for image classification — built a 20-species bird classifier using TensorFlow, Keras, OpenCV, EfficientNetB0, and VGG16.
Graduate-level, systems-heavy data mining course centered on mining massive datasets at scale with Apache Spark (pySpark).
Core areas covered:
- Distributed computing: MapReduce, HDFS-style file systems, Spark RDDs/DataFrames
- Frequent pattern mining: Apriori-style association rule mining at scale
- Similarity search: Shingling, MinHashing, Locality-Sensitive Hashing (LSH)
- Recommender systems: content-based & collaborative filtering, matrix factorization
- Graph mining: social network analysis, PageRank-style link analysis
- Streaming algorithms: Bloom filters, distinct-element counting, moment estimation
- Clustering at scale and web-scale ad-matching
Includes 6 pySpark assignments plus a final data mining competition project (recommendation system, built end-to-end on real large-scale data).
Graduate Deep Learning coursework building neural network fundamentals from scratch rather than treating frameworks as black boxes — automatic differentiation, backpropagation, CNN-based image classification, and physics-informed neural networks (PINNs).
| Assignment | Techniques | Result |
|---|---|---|
| CNN Image Classification | Custom CNN, dropout, training/eval loops | 59% test accuracy on CIFAR-10 |
| Automatic Differentiation | Forward/reverse-mode autodiff | Gradients verified against PyTorch autograd |
| Neural Network from Scratch | Manual backprop, L2 regularization | Gradient checks matched analytical derivations |
| Physics-Informed Neural Networks | PINN, boundary + PDE residual loss | Absolute error 1e-4–5e-3 vs. analytic solution |
Python PyTorch / TensorFlow pySpark scikit-learn SQL Jupyter
Open to data science, ML engineering, and AI/software engineering roles across the Bay Area, LA, and NYC.