Hello :) Today is Day 251!
A quick summary of today:
- finished LingoMate - our project for the Seoul Tech Impact hackathon
I have not slept in more than 24 hours now, so this one will be quite short.
Here is the repo for our project. In here, is all the info and code of our hackathon project.
I also recorded a 1 min video as a demo (which was part of our ppt)
And here is some info from our ppt
What it does
One key to learning a new language is consistent interaction with native speakers and exposure to diverse expressions. LingoMate offers user-friendly, interactive apps that make language learning simple and engaging. Compared to books and video courses, LingoMate is easy to use, simple, and accessible.
How we built it
- Ollama 3.1: One of the LLMs we have used.
- Grafana: Used for monitoring and tracking RAG performance/response.
- ElasticSearch: Used as the vector database.
- Postgres: Used to save user feedback and conversation histories.
- Streamlit: Implemented the interface with limited resources.
Challenges we ran into
- Optimizing prompts to yield appropriate/consistent answers.
- Finding/preparing datasets.
- Enhancing RAG tracing.
Accomplishments that we’re proud of
- I was new to AI, and I am proud that I could absorb/experience LLM in such a short time.
- It was good to put myself in a learner’s shoes (who I used to be) and explore how we can help language learners.
What’s next for LingoMate
- Expand into multiple languages. Offer more combinations of input/output languages to scale it up.
- Scale up datasets, cover broader topics, and enhance model performance.
- Offer features to track histories of conversations for a user.
- Enhance UI.
- With data-driven recommendations, we are aiming to let users personalize their learning.
Finally, below is a pic from the stream when we were presenting:
Time to sleep 😴
That is all for today!
See you tomorrow :)