Portfolio Repository

Python Projects

A comprehensive collection of my work in Data Logic, Optimization, Visualization, and Artificial Intelligence.

Data Structures & Algorithms

RPG Game Demo (OOP Principles)

Python OOP Classes

A playable demo simulating a video game character logic. The character manages statefulness including Health, Magic, and Inventory (Potions). Demonstrates object-oriented design patterns, encapsulation, and method interaction to dynamically modify character stats.

Data Visualization

Special Ed Funding & Outcomes

Matplotlib Pandas EDA

Hypothesis: Enrollment of children with disabilities differs significantly across states and disability categories, indicating varying levels of accessibility and support within special education programs.

Project Overview: This project analyzes the 2022-2023 IDEA Section 618 Child Count dataset to explore enrollment patterns, demographics, and service distribution. The objective is to identify states with the largest served populations and reveal gaps in inclusion.

Key Visualizations:

  • Choropleth maps: Geographic variation in students served under IDEA Part B.
  • Stacked bar charts: Comparison of educational environments across disability types.
  • Box plots: Distribution of enrollment rates and statistical outliers.
  • Bubble charts: Relationships between inclusion rates, total enrollment, and region.
  • Heatmaps: Demographic representation by race, gender, and disability.

Optimization

Production Planning (Linear Programming)

SciPy Optimization Sensitivity Analysis

Goal: Minimize production and outsourcing costs for 5 product types while meeting retailer demand under capacity constraints.

Description: Formulated a Linear Programming model to optimize resource allocation (machining, assembly, finishing). Balanced in-house production with outsourcing to satisfy contractual obligations. Conducted sensitivity analysis to identify process bottlenecks and capacity improvements.

Results: Determined the optimal production schedule to minimize total costs and provided strategic recommendations for capacity expansion.

Machine Learning

Quality Control (Decision Trees)

Sklearn Classification Feature Selection

Goal: Improve quality control efficiency by reducing 30 tests to a key subset while maintaining >96% accuracy in predicting batch quality.

Description: labeled 5000 batches as "Good" or "Poor" based on defect counts. Trained a Decision Tree classifier with cross-validation to identify the optimal number of rules. Analyzed feature importance to eliminate redundant tests.

Results: Achieved ~97% validation accuracy using only 4 specific tests (down from 30) and 5 rules. Significantly reduced testing overhead without compromising quality assurance.

Customer Segmentation (Supervised Learning)

Sklearn SVC Grid Search

Goal: Develop a model to classify 5000+ customers into 3 segments (purchase likelihood) based on 50 potential features.

Description: Evaluated 12 classifiers (including RF, KNN, MLP) using 5-fold cross-validation. Identified a subset of 8 relevant features (x46, x6, etc.) to optimize performance. Fine-tuned a Support Vector Classifier (SVC) with an RBF kernel.

Results: The tuned SVC model achieved 96.20% validation accuracy. The solution successfully labeled new unlabeled customer data for targeted marketing.

Deep Learning

CIFAR-10 Image Classification (CNN)

TensorFlow/Keras Computer Vision Deep Learning

Goal: Build and train a CNN to classify CIFAR-10 images (60,000 color images, 10 classes) with >85% validation accuracy.

Description: Developed in Keras/TensorFlow using data augmentation and dropout for regularization. The dataset includes airplanes, animals, and vehicles. Performance metrics were tracked throughout 40 epochs.

Results: Achieved over 85% validation accuracy. Model architecture and evaluation metrics are documented in the notebook.

Fashion MNIST Image Generator (CGAN)

TensorFlow/Keras Generative AI Conditional GAN

Goal: Train a Conditional GAN to generate realistic grayscale fashion items (trousers, pullovers, sneakers) conditioned on class labels.

Description: Focused on generative modeling using Keras/TensorFlow. Both generator and discriminator received class labels to guide image creation. The model learned to produce visually coherent images aligned with specific categories.

Results: Successfully generated class-specific images with clear differentiation. Notebook includes generated image grids and training progression.

IMDB Sentiment Analysis & Text Gen (LSTM & GPT-2)

NLP LSTM GPT-2

Goal: Classify movie reviews using Bidirectional LSTM and fine-tune GPT-2 to generate realistic, positively-biased reviews.

Description: Explores sentiment classification and generative text modeling. A bidirectional LSTM was trained for binary classification. GPT-2 was fine-tuned to generate positive reviews, which were then validated by the classifier.

Results: The LSTM achieved strong classification performance. The fine-tuned GPT-2 generated coherent positive reviews successfully identified by the classifier, demonstrating a feedback loop.

Contact Me

Get in Touch
© 2026 Steven Lora. All Rights Reserved.