Lingua IC Hack 25

Inspiration

One of our team members faced significant challenges learning English upon arriving in the UK, especially with spoken communication. Traditional language learning relies on static content and dictionary-based methods, which often prove boring and ineffective. Modern platforms like Duolingo adopt these outdated techniques, leaving a gap for a more dynamic, engaging approach.

What it does

Lingua leverages an Anki-based backend for active recall and uses generative AI to create dynamic, real-time conversations that incorporate targeted vocabulary cards. It adjusts the stability (frequency of appearance) of each card based on user responses and employs an InterSystem vector database search to identify similar words in the card bank, optimising learning efficiency and speed by adjusting the stability of these words too.

How we built it

We built Lingua by integrating FastAPI for the web backend, OpenAI's APIs for generative conversation, speech synthesis, and transcription, and an IRIS vector database for similarity searches. The code ties together active recall techniques with state-of-the-art AI to deliver an adaptive and interactive learning experience.

Challenges we ran into

Our journey included dependency issues, confusing API documentation, and the classic “works on my machine” dilemma. These challenges pushed us to innovate and troubleshoot under pressure, ultimately strengthening our resolve and teamwork.

Accomplishments that we're proud of

We are incredibly proud of the bond we forged during this project despite limited initial experience with generative AI tools. Overcoming technical hurdles under tight time constraints, we collaborated effectively and built a prototype that truly reflects our passion for solving language learning challenges.

What we learned

The modern language learning industry is massive, Duolingo is valued at around $16 billion, but many existing solutions do not maximise learning efficiency. Our experience shows that by combining active recall, generative AI, and similarity searches, we can create a more engaging and effective learning platform that keeps users immersed for longer.

What's next for Lingua

With our MVP/prototype complete, our next steps involve scaling the platform for broader use and refining its accuracy, especially addressing the challenges in speech-to-text conversion for similar-sounding words. Ongoing research and development in these areas will be crucial to making Lingua a disruptive force in the language learning space.

Built With

• css
• fastapi
• html5
• intersystem
• intersystems
• iris
• javascript
• langchain
• openapi
• python
• websockets