What you get
- Cleaned, combined datasets you can trust and reuse.
- Exploratory analysis that surfaces trends, outliers, and the drivers behind them.
- Predictive models for forecasting, classification, and customer segmentation, with plain-English explanations.
- Interactive dashboards and charts built for the people who will use them.
- Documented, repeatable code, so the analysis can be rerun next month with fresh data.
Modern data science techniques I use
Fast analytics without a server, using DuckDB and Polars
Not long ago, analyzing tens of millions of rows meant standing up a database server. Today, embedded engines like DuckDB and DataFrame libraries like Polars query CSV, Parquet, and even Excel files directly, at speeds that rival a data warehouse. For small and mid-sized teams, that means faster answers at a fraction of the cost.
Columnar formats and the open lakehouse
Storing data in Apache Parquet, with Apache Arrow for in-memory exchange, shrinks file sizes and speeds up queries. Open table formats such as Apache Iceberg and Delta Lake add versioning and safe updates on top, so one copy of your data can serve Python, SQL, and BI tools at the same time.
Analytics engineering with dbt and Git
Data transformations belong in version-controlled SQL, not in a chain of hand-edited files. Tools like dbt let me write tested, documented models, and automated data quality tests catch broken source data before it reaches a report.
Explainable machine learning
For tabular business data, gradient-boosted trees such as XGBoost and LightGBM remain the workhorses. I pair them with SHAP values, which show which factors pushed each prediction up or down. A model you can question is a model you can trust.
AI-assisted analysis, used carefully
Large language models speed up work like categorizing free-text survey comments, grouping similar records with text embeddings, and drafting queries. I treat their output as a first draft and check it against the source data every time.
Visualization that tells the story
I build interactive web charts with Plotly, Vega-Lite, and D3.js, and dashboards in Power BI or Tableau when your team already works there. The chart type follows the question: trends get lines, comparisons get bars, distributions get histograms, and nothing gets a 3D pie chart.
Tools and technologies
- Python
- pandas
- Polars
- DuckDB
- SQL
- PostgreSQL
- scikit-learn
- XGBoost
- SHAP
- dbt
- Jupyter
- Plotly
- D3.js
- Power BI
- Tableau
- Git
Common questions
How much data do I need before data science is worthwhile?
Less than most people expect. A clear question matters more than volume, and a few thousand clean rows can answer a lot. For predictive models, I'll tell you upfront if your data isn't enough to produce reliable results.
Can you work with data we already keep in Excel?
Yes. Spreadsheets are the most common starting point. I can analyze them as they are, or clean and combine them first so future analysis is easier.
Will I be able to update the analysis myself?
That's the goal. You'll get either a dashboard that refreshes from your sources or documented scripts with simple instructions for rerunning them.