Data Analyst · beginner
Laptop Lakehouse: DuckDB, Parquet, and a One-Page Dashboard
A complete analytics project that runs on your laptop for free — public data converted to Parquet, queried with SQL in DuckDB, modeled into clean metrics, and shipped as a one-page dashboard you can actually send to someone.
Project overview
DuckDB plus Parquet has become the default starter stack for analytics in 2026 — it is what analysts use before a company buys a warehouse, and what analytics engineers use to prototype. Finishing a clean project end to end matters more to hiring managers than any single tool.
Real-world use case
A city open-data portal publishes millions of rows of service requests as CSV. You turn it into a fast, queryable dataset and a dashboard that shows response times by neighborhood.
Tools and technologies
- DuckDB
- Python
- Parquet
- Streamlit or Evidence
- SQL
Skills you can demonstrate
SQL Analytics, Columnar Data Formats, Data Modeling, Dashboard Design, Data Storytelling
Build roadmap
- Pick a real dataset
Choose a public dataset with at least a million rows and a date column — city service requests, transit ridership, weather. Download the raw files. - Load and convert
Read the CSV with DuckDB, fix types, and write partitioned Parquet. Compare file size and query speed before and after — note the numbers. - Explore with SQL
Answer ten questions with SQL: totals, trends, top-N, distributions. Save each query. Notice where the data is dirty and document it. - Build clean metric tables
Write SQL views for the three or four metrics the dashboard will show. Every number on the dashboard should trace to one view. - Ship the dashboard
One page, four visuals, one clear headline. Use Streamlit or Evidence. Deploy it free so the link works in an application. - Write the findings
A short README: what the data says, one surprising thing, and what you would look at next. This is the part a hiring manager reads.
Ways to extend the project
Publish the dataset as a set of Parquet files on a free static host and query them remotely from DuckDB over HTTP — a "serverless data lake" in one afternoon.
How to showcase your work
- GitHub: The notebook or scripts, the SQL models, the dashboard app, and a README with two screenshots and one paragraph of findings.
- LinkedIn: Post the dashboard image with three sentences of insight. Analysts get hired for the insight, not the chart.
- Resume: Frame it around outcomes: "Modeled 3M public records with DuckDB and Parquet and delivered a dashboard that surfaced a 40% backlog gap between districts."