Umi

My co-founder and I set out on a mission to make it possible to become fluent in a foreign language without moving to a foreign country.

We wanted to solve one of the core problems in language learning, which is that many language learners are unable to understand native speech after taking a course or using an app. You may be familiar with this problem if you've ever taken high school Spanish/French/etc in the USA or used Duolingo.

In both mine and my cofounders' experiences, we'd discovered that the best way to gain comprehension ability is via immersion learning, which is inspired by the Input Hypothesis.

We built an NLP/LLM processing pipeline to turn real world videos into bite-sized language lessons. These lessons helped learners gain multiple skills at once, including listening comprehension, pronunciation, and vocabulary.

From a technical perspective, LLMs helped us solve problems that were previously expensive to solve at scale such as clipping videos semantically, validating subtitle accuracy, explaining words & phrases, sentence separation (Japanese, Chinese, etc), and contextual definitions.