Blendin

Inspiration

Frustration with current localization flows. They're high-friction, slow, and expensive. Opportunity with LLMs to make localization flows seamless, fast, and cheap.

What it does

Hackathon Project: Take a website url, download the html content of the website, extract the strings from the page, translate the strings to multiple languages, re-inject the strings and host the website to a remote url

Ultimately: A suite of developer tools to integrate automated localization flows into web and mobile application code and devops processes.

How we built it

Terraform and AWS to run the data pipeline. Ruby code running on Lambda functions to do the HTML parsing and API calls to OpenAI's gpt-4 API for language translations.

Challenges we ran into

• Rate-limiting by OpenAI's API

• Downloading a full website can be quite heavy, as websites often have many horizontal links and can host large files on their domain

• XML parsing library was using native libraries, and erroring on AWS Lambda runtime. Had to make a Docker container to model AWS Lambda and compile on that.

Accomplishments that we're proud of

Having a working demo in 24 hours.

What we learned

Learned how other people in the community are leveraging different AI tools

What's next for Blendin

Getting initial user base of simple localization integrations for indie developers who aren't currently localizing their apps. (secured first user at the hackathon) Get to ~$2,000 MRR bootstrapping within own network, then assess how best to scale up the app

Built With

• amazon-web-services
• ruby
• terraform