# Apache Parquet Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/apache-parquet
> Markdown URL: https://aitinkerers.org/technologies/apache-parquet.md
> Technology record last updated: 2026-03-13T13:39:17Z
> Generated: 2026-08-24T11:50:14Z

Apache Parquet is an open-source, column-oriented data file format optimized for efficient data storage and high-performance analytical query processing.

Parquet is the definitive columnar storage format for big data analytics: it drastically improves performance and reduces storage costs. Unlike row-based formats (CSV), Parquet organizes data by column, enabling systems to read only the necessary fields (predicate pushdown) and compress similar data types together (e.g., Snappy, Gzip). This design delivers massive efficiency gains; for example, one study showed Parquet queries running 34x faster with 99.7% cost savings compared to CSV on a 1TB dataset. The format is language-agnostic and features a self-describing schema via a file footer (metadata), making it the standard interchange format across major ecosystems like Apache Spark, Hive, Presto, and cloud services (AWS Athena, Google BigQuery).

- Official technology site: https://parquet.apache.org/
- Public AI Tinkerers demos and talks: 2
- Result page: 1 of 1

## Recent Public Talks and Demos

### [How We Put a Data Warehouse in the Browser and Let AI Agents Explore It](https://toronto.aitinkerers.org/talks/rsvp_j4OM0Xw7uFQ)

We built a customer-specific data warehouse that runs entirely in the browser. At Relay, every customer has a personalized financial dataset containing transactions, balances, and other banking data. Traditionally, exposing insights from that data required backend APIs, analytics services, and purpose-built dashboards. Instead, we export customer data into compressed Parquet files, load them into DuckDB-WASM, and run a full analytical SQL engine directly in the browser. Every customer effectively gets their own tiny data warehouse. The surprising part came next: once the warehouse existed locally, AI agents became dramatically simpler. Instead of building tool chains, APIs, or MCP servers, we gave agents direct SQL access to the customer's warehouse and let them investigate the data themselves. In the demo I'll show: - A customer-specific warehouse running entirely in the browser - Analytical queries executing locally with no backend round trips - AI agents exploring financial data and generating insights - The architecture powering this in production - The code behind the browser warehouse and agent workflows - The challenges we hit scaling hundreds of tiny warehouses

- Event context: AI Tinkerers Toronto - June 2026 - with Relay Financial x PostHog — 2026-06-30 — Toronto
- Public talk page: https://toronto.aitinkerers.org/talks/rsvp_j4OM0Xw7uFQ

### [LLM Datasets in the Browser with Parquet](https://seattle.aitinkerers.org/talks/rsvp_EO-LDz-ikX8)

6 months ago I started looking at LLM datasets on huggingface. Most of them are in Apache Parquet format. I wanted to LOOK at the data in these files. And of course it is possible to query parquet in duckdb or a notebook, but if you want to view it in a modern UI, the browser, there was no good way to do so. So I build Hyparquet - an open source Apache Parquet parser written in pure JavaScript, which can stream data from remote urls. I will talk about how I built it, why it's useful, and give a demo of loading massive (multi-gigabyte) datasets in the browser, with no backend.

- Event context: AI Tinkerers Seattle - June 2024 — 2024-06-07 — Seattle
- Public talk page: https://seattle.aitinkerers.org/talks/rsvp_EO-LDz-ikX8

## Related Technologies

- [Apache Arrow](https://aitinkerers.org/technologies/apache-arrow) ([Markdown](https://aitinkerers.org/technologies/apache-arrow.md)) — 1 public demo
- [DuckDB](https://aitinkerers.org/technologies/duckdb) ([Markdown](https://aitinkerers.org/technologies/duckdb.md)) — 9 public demos
- [DuckDB-WASM](https://aitinkerers.org/technologies/duckdb-wasm) ([Markdown](https://aitinkerers.org/technologies/duckdb-wasm.md)) — 1 public demo
- [Hugging Face](https://aitinkerers.org/technologies/hugging-face) ([Markdown](https://aitinkerers.org/technologies/hugging-face.md)) — 41 public demos
- [Hyparquet](https://aitinkerers.org/technologies/hyparquet) ([Markdown](https://aitinkerers.org/technologies/hyparquet.md)) — 1 public demo
- [JavaScript](https://aitinkerers.org/technologies/javascript) ([Markdown](https://aitinkerers.org/technologies/javascript.md)) — 40 public demos
- [Python](https://aitinkerers.org/technologies/python) ([Markdown](https://aitinkerers.org/technologies/python.md)) — 654 public demos
- [React](https://aitinkerers.org/technologies/react) ([Markdown](https://aitinkerers.org/technologies/react.md)) — 215 public demos