Google rolls out new dev tools focusing on open source and GenAI

Google on Wednesday rolled out a series of GenAI-powered developer tools, including a coder-focused translation tool that one IDC analyst described as “an absolutely amazing and remarkable project.” 

Language translation is nothing new, but the particular challenges of translating coding phrases into multiple languages while allowing the GenAI engine to understand the nuances of intent and accurately conveying them in multiple languages is quite difficult, said Arnal Dayaratna, the IDC vice president for software development.

Among the various tools introduced by Google Wednesday were: an expansion of Project Vanni, Google’s work with the Indian Institute of Science (IISc) “to capture the diversity of India’s spoken languages”; IndicGenBench, a comprehensive benchmark designed to evaluate generation capabilities of LLMs on Indic languages; and CALM, a Composition of Language Models framework that Google said should help coders “to combine your specialized language models with Gemma models.”

Google also introduced the MatFormer framework, which Google said is supposed to help developers to “mix and match different AI models within a single framework, optimizing for both high performance and low resource consumption.”

Project IDX, an AI-assisted integrated workspace for full stack developers which now features new templates for backends and databases, was moved on Wednesday to public beta. 

The Google rollouts also included Project Oscar, a reference architecture for an AI agent that helps with open source project maintenance that Google said it is now going to release as open source.

IDC’s Dayaratna said he was particularly intrigued by Project Vanni because it directly addresses a particularly difficult coding obstacle, especially when development teams exist in many different countries speaking many different languages. The problem can also exist in countries with many different local languages, such as when widespread Indian teams have speakers writing in, for example, Hindi, Bengali, Marathi and Kannada.

The Vanni project is dealing with two difficult challenges. One is translating coding references without missing key technical details, especially when it is being translated into multiple languages. The second, potentially more severe, problem is doing these translations within a GenAI environment, where the algorithm can hallucinate or even simply try to extrapolate meaning where it might predict the next word incorrectly. If that happens, the translations become confusing and possibly destructive.

“What (coders) need is translation that is contextual and is specific to natural language prompts in software,” Dayaratna said, adding that translation specialized for developers “does not exist” yet. 

He gave an example of a coder saying that he wants to create a cartesian product for specific variables. “There is a strong likelihood of mistranslation because Google doesn’t understand what cartesian is,” Dayaratna said.

Vanni also “directly addresses the challenge of enabling collaboration. We see something very special with this project,” Dayaratna said.

Project IDX is also interesting because of its focus on transitioning dev tools to code environments where a much higher percentage is in the cloud. This change has the potential to greatly accelerate coding, but it needs higher performing compute resources. In theory, IDX would allow coders “to instantly provision the dev tools that they need, and use those to get work done faster. They don’t have to wait for IT to install those tools on their machines.”

IDX also should help a different kind of language translation, by delivering help moving between programming languages such as Javascript or Python. “It has preconfigured templates,” Dayaratna said.

Project Oscar also has interesting potential, Dayaratna said, and “is remarkable because AI agent technologies are embryonic at present. AI agents that are open source is absolutely remarkable. Google is demonstrating an increased commitment with respect to open source and developer tools, compared with their prior commitments.”