IntellEco Translate

The problem

Over 1 billion people use online translation tools to break language barriers and billions of translation requests are served every day (ref. Google Translate). The current translation services are not carbon efficient and most users do not know how they can help reduce carbon emission when using translation services.

Our solution

IntellEco Translate informs users about the carbon footprint of their translation requests and provides the “Eco-mode” choice, which will reduce the carbon footprint by serving users' requests in the most carbon efficient region.

How IntellEco uses Carbon Intensity Data and SDK?

IntellEco Translate uses both the historical carbon intensity data provided by Watttime and the real-time carbon rating data provided by the SDK. 

First, IntellEco Translate analyzed the historical carbon intensity data from 2018 to 2022 and discovered five regions (France Central, West US, East US, Australia, and UK South) to deploy our services in the Azure cloud. These five regions are carefully selected because: 

(1) They provide large geospatial coverage in North America, Europe, and Asia; 

(2) They have historically low carbon values;

(3) Azure supports the VM deployment in these regions.   

When users choose Eco-mode, the SDK is queried to get the real-time carbon rating for each selected region and send the user request to the region with the lowest carbon rating. IntellEco Translate also notifies users of their carbon savings in percentage format.

What have we achieved during the hackathon?

Through the hackathon, we leveraged both the historical and real-time carbon intensity data provided by the SDK and built the fully working demo deployed on five regions of the Azure cloud.

The impact of our solution

Google Translate alone is used by over 1 billion users and translates around 100 billion words every day. Assuming 1 billion translation requests are served each day, IntellEco Translation can save approximately 30,000 tons of CO2 per year. In addition, there are many other online translation services such as translating the entire web page and translation services embedded in various social media apps. If all these translation services can use carbon aware IntellEco Translation, hundreds of thousands of tons of CO2 emissions can be reduced every year.    

The other equally important impact is that people from all over the world who speak different languages will learn the impact of user choices (e.g. switch to Eco-mode) on carbon reduction. We believe IntellEco Translate will inspire millions of users on promoting green software design and carbon aware usage.

Feasibility

IntellEco Translate has been deployed on Azure cloud in five different regions (France, East US, West US, UK, and Australia). This allows many users from around the world to try our demo out and see the impact of carbon aware translation. Since IntellEco Translate is a cloud based solution, it can easily scale up by deploying more servers at different regions when the number of requests increases. 

Vision and road map

Currently, IntellEco Translate aims to find the “global minimum” of carbon rate. Although it maximizes carbon savings, it could sacrifice the response time as well. To better serve users from different countries, a compromised solution would be to find a “regional minimum” carbon rate with least impact on response time. To reduce cost, IntellEco Translate is deployed on low cost CPU VMs and uses free software solutions. The prize (if awarded) will allow us to (1) deploy IntellEco Translate on GPU VMs for faster response time and more carbon savings and (2) resolve the issues of using free software solutions (e.g. temporary time we can use the service, watermark etc.).   

Our vision for IntellEco is beyond online translation. The framework we developed during this Hackathon can be used to deploy many AI models to support various carbon aware AI services (e.g. the chatbot used in automatic custom services, the audio to text or text to audio services etc.).

Prototype URL and GitHub repositories

• Link to prototype
• Link to Github repository