Hi! I’m Corey, a Python developer and ML engineer with a foundation in education, automation, and research-based AI systems. I work across disciplines to build reliable tools and intelligent workflows for solving real-world problems.
My projects span:
- Applied machine learning in NLP, classification, and reinforcement learning
- Automation and data pipelines using modern scraping and API tools
- Interactive applications for education, research, and end-user analysis
- Research replications of academic ML papers with scalable results
A faithful implementation of “Pokémon Red via Reinforcement Learning” (arXiv:2502.19920), training an agent to reach Cerulean City using PPO + LSTM in a long-horizon, multi-reward setting.
Highlights:
- 50M+ steps of training with hybrid observations (RAM + visual)
- Multi-component reward shaping + animation fixation detection
- Academic PPO configuration with milestone-based tracking (Mt. Moon, Pewter Gym, etc.)
- Full training logs via TensorBoard and modular config system
🧪 Reproduces research-level performance in a high-complexity gaming environment.
A fine-tuned BERT model with a custom dual attention mechanism (DualBert) to classify educational questions by grade level (3rd–12th).
Highlights:
- Inspired by Song et al.'s attention modeling architecture
- Trained on QxGrade Dataset
- ~70% accuracy on held-out educational questions
- Deployable via Streamlit for real-time inference
- Built with PyTorch, Hugging Face, and Google Colab for training
📘 Future development includes Bloom's Taxonomy tagging and deeper complexity modeling.
An end-to-end ETL pipeline for scraping and analyzing state-level education funding from Excel spreadsheets published across U.S. education departments.
Highlights:
- Scrapes files using Playwright and automates uploads to Google Drive
- Consolidates multi-tab spreadsheets with Google Apps Script
- Outputs clean pivot tables by state in Google Sheets
- Asynchronous I/O with secure Google OAuth2 integration
- Designed for transparency, maintainability, and rapid use by analysts
📈 Enables policy-facing insights into state-level education grant distributions.
A curated dataset of Common Core-aligned educational questions labeled by grade level (3–12), designed to support text classification and ed-tech research.
- Labeled by pedagogical level
- Suitable for training and evaluation of NLP models
- Used in the
Studybotproject above
Languages: Python, SQL
ML & NLP: PyTorch, Hugging Face Transformers, scikit-learn
Data: pandas, NumPy, openpyxl, Google Sheets API
Automation: Playwright, Selenium, Google Drive API, Google Apps Script
Apps & UI: Streamlit, Jupyter, Google Colab
Monitoring & Ops: TensorBoard, GitHub Actions
| Project | Description | Key Tools |
|---|---|---|
| PokeBot | RL agent for Pokémon Red game | PPO, LSTM, Gym, TensorBoard |
| Studybot | Classifier for grade-level question difficulty | BERT, Dual Attention, Streamlit, Colab |
| GrantFundingPipeline | ETL pipeline for state education grants | Playwright, Google APIs, Excel, Colab |
| QxGrade Dataset | Grade-labeled question dataset (3–12) | Kaggle, CSV, CCSS-aligned annotation |
- 📧 Email: [email protected]
- 🔗 GitHub
All repositories follow open-source or fair-use licensing. Please cite original sources where applicable, especially for research-based implementations such as PokeBot.