Selected Projects
A few representative projects across forecasting, warehousing, analytics apps, and automation.
Kaweah River LSTM Flow Forecasting (Time-Series ML)
Python
TensorFlow/Keras
scikit-learn
Developed and evaluated Long Short-Term Memory (LSTM) neural networks to forecast hourly river discharge for a steep, snowmelt-driven watershed,
supporting operational decision-making for commercial whitewater rafting.
Built a pipeline using 18 years of hourly flow data and multi-elevation SWE measurements to predict river levels up to two weeks in advance.
Key contributions
- Assembled and cleaned multi-source hydrologic datasets; domain-informed imputation for missing data
- Time-aware train/validation splits to prevent leakage; univariate + multivariate LSTMs
- Benchmarked vs persistence using RMSE, MAE, and Nash–Sutcliffe Efficiency
- Diagnosed peak snowmelt failure modes; identified missing climate inputs as primary constraint
Outcome
Models learned seasonal & recessionary patterns and beat persistence in stable regimes.
Clear next step: incorporate temperature/climate drivers for peak melt events.
Eco Essential Enterprise Data Warehouse
Snowflake
dbt Cloud
Fivetran
Tableau
AWS
Designed and implemented an end-to-end enterprise data warehouse and analytics pipeline for a simulated eco-friendly cookware company, acting as a data management consultant.
Built a scalable dimensional model to support sales analytics and marketing attribution, integrating e-commerce transactions with email engagement.
Key contributions
- Enterprise dimensional model with bus matrix & conformed dimensions
- ELT via Fivetran from Postgres (RDS) + S3 into Snowflake
- dbt facts/dimensions with documentation and production-style modeling
- dbt tests (unique, not_null, accepted_values, relationships) across models
- Scheduled syncs + dbt Cloud jobs; Tableau dashboard connected to Snowflake
Outcome
Delivered a functioning analytics stack spanning ingestion, transformation, testing, scheduling, and visualization — mirroring real-world warehouse workflows.
Cache County Real Estate Analytics Dashboard + API
Tableau
AWS (S3, Lambda, API Gateway)
Python
REST API
Created a publicly available Tableau dashboard hosted on AWS and a supporting AWS-based API to explore 10 years of Cache County, Utah residential MLS listing data,
leveraging prior experience as a licensed real estate agent to design market-aligned metrics.
Cleaned and transformed 21,054 MLS records using Python, generating 176 JSON datasets by ZIP code and year.
The API enables programmatic access; the dashboard visualizes median prices, acreage, days on market, and value distributions.
Automated Stock-Trading Platform
Python
AWS EC2
AWS Lambda
cron
Alpaca API
Developed an end-to-end Python trading framework that ingests real-time market data, executes Mean Reversion, SMA, and Bollinger Bands strategies,
and tracks daily performance across 10 equities. Implemented short selling, fractional-share support, and dynamic order sizing with error-handling safeguards.
Deployed on AWS EC2 with scheduled execution via cron, automatic logging, and JSON-based result storage; used AWS Lambda to control instance start/stop timing.
Loan Default Risk Modeling (FinTech)
Python
scikit-learn
XGBoost
TensorFlow/Keras
Built and compared supervised learning models (Logistic Regression, tree ensembles, KNN, MLP) on a 307k-record imbalanced loan dataset
to predict borrower default for an inclusive-lending startup.
Engineered features, handled class imbalance, and evaluated with PR-AUC, ROC-AUC, and F1.
Gradient Boosting achieved top performance (PR-AUC = 0.246, ROC-AUC = 0.753) with interpretable feature importances.