LangBase

Inspiration

I am pursuing my Masters in Computational Linguistics and I've really enjoy learning about low resource languages. I like the idea of any language being learnable by anyone and preserving peoples culture. The hardest part of data collection is the manual work so we can let agents do the tedious work.

What it does

Our main focus was parts of speech tagging, translation and text extraction from images. A user can upload data, do specific research on the task and AI will suggestions for parts of speech of that sentence, translate it or extract the text from the image. A user has final say if the suggestion is good. They can deny, accept and update.

How we built it

We use Claude to do research on the grammar and syntactic rules of a language. Browserbase powers Claude with the correct tools to get clean data from websites. We then summarize those into notes and use them to give suggestions on translations and parts of speech. We use previous user choices to make future decisions.

Challenges we ran into

Wifi. Finding data for our chosen sample languages. We had an idea to simulate low resource languages using a small subset for English but then realized Claude would have a very easy time with any task we gave it.

Accomplishments that we're proud of

We got some AI suggestions that seem very promising. It seems that we could scale this to be more research heavy.

What we learned

Learned about browserbase and its many helpful uses cases for giving our agents valuable data.

What's next for LangBase

Not sure maybe we can implement this further and train some dedicated models for translation, parts of speech tagging and character extraction. At the very least a more clean implementation of this project.

Built With

• anthropic
• arize
• browserbase
• fastapi
• react
• supabase