The ggsql Project, GPU Accelerated with Python 3 and CUDE | Issue 85

This week's agenda:

• Open Source of the Week - The ggsql project
• New learning resources - the future of forecasting, setting personal ChatGPT with Open WebUI, Docker, and Gemma 4, yfinance crash course
• Book of the week - GPU-Accelerated Computing with Python 3 and CUDA by Niels Cautaerts and Hossein Ghorbanfekr

The newsletter is also available on Substack and Medium.

Want to learn how to build SQL AI agents in production? I’ll be presenting next week at Maven’s Lightning Lessons on SQL AI agents in production, hosted by Aishwarya Srinivasan and Arvind Narayanamurthy . For more details and to RSVP:

Open Source of the Week

This week's focus is on the ggsql project. ggsql is an open-source project from Posit PBC that brings data visualization directly into SQL by extending standard SQL syntax with grammar-of-graphics concepts such as layers, scales, facets, and labels. Instead of switching from SQL to Python or R for chart creation, users can query data and define visualizations in the same language. This makes analytics workflows more streamlined, reproducible, and accessible—especially for teams that already work heavily with databases and SQL. ggsql is currently in alpha and supports environments such as Jupyter, Quarto, VS Code, and Positron.

Key Features

• Combines SQL queries and chart specifications in a single declarative syntax using clauses like VISUALISE, DRAW, and SCALE
• Inspired by the grammar of graphics, enabling layered and composable visualizations such as scatterplots, histograms, boxplots, and faceted charts
• Pushes computations to the database engine, making it suitable for large datasets without extracting full tables locally
• Supports current backends, including DuckDB and SQLite, with plans for broader database support
• Available across multiple tools, including Jupyter notebooks, Quarto documents, CLI, VS Code, and Positron
• Includes a Python package for integrating ggsql charts into Python workflows and rendering outputs such as Altair charts
• Designed to be readable for both humans and AI agents, making it easier to generate, inspect, and modify chart queries programmatically

More details are available in the project documentation:

License: MIT

New Learning Resources

Here are some new learning resources that I came across this week.

The Future of Forecasting

I had the pleasure of speaking with Richard Cotton at the DataCamp podcast about the future of forecasting.

Setting up Private ChatGPT with Open WebUI, Gemma4, and Docker

I created a short tutorial on setting up a private ChatGPT-like service using Open WebUI, Gemma 4, and Docker, with a few simple steps. No prior Docker knowledge is needed. The tutorial is available on Medium:

And in my AIOps newsletter:

yfinance Crash Course

The following tutorial by NeuralNine introduces the yfinance Python library. This library provides a Python client to Yahoo! Finance’s API.

Book of the Week

This week's book focus is on GPU programming with CUDA - GPU-Accelerated Computing with Python 3 and CUDA by Niels Cautaerts and  Hossein Ghorbanfekr . This hands-on book teaches Python developers how to harness GPU computing for high-performance workloads using NVIDIA CUDA and the modern Python ecosystem. It bridges low-level GPU programming with high-level productivity tools, showing how to speed up scientific computing, data processing, and machine learning workflows without switching to C++. The book combines core CUDA concepts with practical examples, performance tuning, and real-world GPU applications with Python libraries such as JAX, CuPy, RAPIDS, and Numba.

Topics covered

• CUDA fundamentals in Python — understanding GPU architecture, parallel execution, and writing custom kernels with Numba-CUDA.
• Debugging and profiling GPU code — using NVIDIA Nsight tools to analyze performance bottlenecks and optimize execution.
• Memory optimization — efficient memory access patterns, host/device transfers, and maximizing throughput.
• Asynchronous execution — CUDA streams and overlapping computation with data movement for faster pipelines.
• Multi-GPU scaling — distributing workloads across GPUs using tools like Dask-CUDA.
• High-level GPU Python libraries — accelerating workflows with JAX, CuPy, RAPIDS, and Numba.
• Scientific computing use cases — GPU acceleration for PDE solvers, simulations, and numerical workloads.
• AI and deep learning workloads — building and running transformer models on GPUs.

This book is ideal for Python developers, data scientists, engineers, and researchers who already use tools like NumPy, Pandas, or SciPy and want to significantly improve performance through GPU acceleration while staying in the Python ecosystem.

The book is available for purchase on the publisher’s website and on Amazon.

Have any questions? Please comment below!

See you next Saturday!

Thanks,

Rami

📌 P.S. I share daily updates on Substack, Facebook, Telegram, WhatsApp, and Viber.