Meet This Indian Developer Who Built a JupyterLab-Killer

In the past few days, Hacker News has been abuzz about the new ‘JupyterLab-killer’ called Zasper. Within a few days of its release, the latest open-source tool has garnered widespread attention and praise from industry experts, including JupyterLab’s creators.

Why Zasper? 

Hyderabad-based developer Prasun Anand created Zasper, the open-source alternative to JupyterLab – a tool widely used by data scientists. Zasper, vaguely derived from ‘Jupyter as a service,’ is said to deliver significant improvements in performance and efficiency. 

Despite its popularity, JupyterLab has been criticised for its slow performance and the tendency to hang when running multiple notebooks. Anand’s solution tackles these problems head-on by reimagining the software’s architecture.

Zasper’s backend uses the Go programming language instead of Python. This change allows it to take advantage of multi-core processing, a significant improvement over JupyterLab’s single-core limitation. 

“Go is a better platform. When I wrote the first draft, it automatically became 4x better than Jupyter,” said Anand in an exclusive interaction with AIM. 

While JupyterLab typically consumes around 120 MB of RAM, Zasper uses only 25-27 MB. CPU usage shows a similar improvement, with Zasper using 0.2 CPUs compared to JupyterLab’s 0.8 CPUs. These gains are significant, considering JupyterLab is an essential tool in data science workflows. 

“JupyterLab is like a toothbrush for a data scientist,” Anand explains. “They have to use it every day to run basic code. If a data scientist works for eight hours, he will stare into JupyterLab for four hours, even five.”

Although JupyterLab is the preferred choice for most data scientists, it has several shortcomings. It has been criticised for performance issues, particularly when handling large datasets or multiple notebooks simultaneously. 

Furthermore, it faces challenges in code structuring, production readiness, and security. In production, its non-linear workflow and lack of robust error handling hinder reproducibility and scalability. 

What About the Other Alternatives? 

Some worthy opponents to JupyterLab or Zasper are Databricks and DeepNote. In fact, DeepNote has been backed by powerful names, such as OpenAI’s Greg Brockman and entrepreneur Naval Ravikant, as well as VC firms like Accel and Index Ventures. 

However, the notebook solutions developed by these players remain closed-source. This is where Zasper has an edge over the others. Anand’s decision to make Zasper open-source aligns with the collaborative spirit of the data science community and could accelerate its development and adoption. 

Anand said that Zasper’s success highlights the importance of individual contributors in the open-source ecosystem. “This is how open source works, right? Someone comes up with a crazy idea, and then they build it. This is how Linux was born. And even Jupyter was born in this way,” he remarked. 

Journey Thus Far

Anand’s journey to creating Zasper is rooted in his extensive experience with open-source software development. A chemical engineering graduate, he became involved in coding during his college days at BITS Pilani. His background includes participation in Google Summer of Code, contributions to scientific software for Ruby, and work on Facebook’s PyTorch library.

Zasper’s development began in June 2023, and Prasun has been working on it full-time for the past seven months. He released an initial version in September and continues to build it publicly to encourage discussion and contributions. 

The project reached a usable state on December 27, 2023, and gained widespread attention in early January 2024. 

While Zasper shows great promise, Anand acknowledges that it’s still in the minimum viable product stage. “It needs further improvement to make it better,” he said. “Once you have users, you get their feedback and work on the critical features.” 

Its current limitations include the need for more kernels, such as SQL and Julia, and support for Linux systems. He is also working on improving features like file searching.

Anand estimates that the software will become fully stable in about two years. He also confirmed that he was in talks with a few VC firms and hoped for something to materialise.