Nickmoon (Moon) Mware

Context

Public financial-services analytics project built from CFPB complaint data, with careful caveats around representativeness and responsible interpretation.

Key insight

Complaint data is most useful as a monitoring signal when it is paired with product context, geography, response outcomes, and clear data limitations.

Business impact

Created a reusable analytics workflow that helps financial-services teams identify which product and issue patterns deserve deeper operational review.

Problem

Financial complaint data is large, messy, and easy to misread without a focused workflow for cleaning, filtering, aggregating, and explaining the limits of the dataset.

Methods

Processed the CFPB CSV ZIP locally, filtered financial product categories, cleaned dates and categorical fields, stored analysis-ready data in SQLite, and generated dashboard metrics for product, issue, state, and response patterns.

Findings

The initial sample highlights credit reporting and debt collection as high-volume monitoring areas, with Texas complaint patterns available for regional comparison and follow-up analysis.

Context

Framed as a customer analytics case study for an eBook store, focused on helping non-technical stakeholders understand who is most likely to subscribe and spend more.

Key insight

The strongest value comes from turning model output into targeting decisions, not just reporting accuracy scores in isolation.

Business impact

Built customer spend and subscription models using PyCaret and gradient boosting so marketing and sales teams could prioritize higher-propensity customer segments with more confidence.

Problem

The business needed a clearer way to identify customers likely to subscribe and estimate monthly value before spending budget on broad outreach.

Methods

Prepared customer features, compared multiple regression and classification approaches, and selected models that balanced predictive performance with stakeholder interpretability.

Findings

The final workflow highlighted which customers were more likely to subscribe and spend more, giving the team a more targeted lens for campaign planning and customer prioritization.

Context

Conceptual and research-backed educational AI project shaped by teaching experience, chatbot design thinking, and current work in applied analytics.

Key insight

The most effective learning assistants do more than answer questions. They support comprehension, reduce friction for students, and align responses with instructional goals.

Business impact

Designed an educational chatbot concept using instructional context and NLP framing so student support could become more scalable, consistent, and accessible outside normal class time.

Problem

ASR systems often struggle with speaker variability, producing higher transcription error rates for different accents and speaking styles.

Methods

Processed WAV audio from Wikimedia Commons, generated mel spectrograms, trained a CNN-based speech workflow with Python libraries such as Librosa, NumPy, and PyTorch, and evaluated performance using WER.

Findings

The trained model improved transcription accuracy from roughly 27% to about 77%, while the paper outlines future work on accent robustness, MFCC features, and broader speech datasets.

Context

Exploratory healthcare analytics project on a readmission dataset that hiring managers will immediately recognize.

Key insight

This project is strongest when it shows the full workflow: profiling, missing data treatment, feature behavior, and a clear next modeling step.

Business impact

Analyzed readmission patterns and data quality issues so the dataset could support more reliable risk modeling and operational decision-making in a healthcare context.

Context

Public dashboard work intended for recruiters, hiring managers, and non-technical stakeholders reviewing projects.

Key insight

Each dashboard should communicate the business question, dataset context, and primary takeaway within the first few seconds.

Business impact

Built story-first dashboards in Tableau so viewers could move from chart reading to decision insight quickly, without needing a live walkthrough.