Read Panda

• 

  wontons, yum!

• 

• 

• 

• 

Inspiration

A lot of kids are falling behind in reading due to the pandemic. Yi's classmate Cindy started an amazing student to student tutoring program with the Kiwanis Club Fairfield that tutored 193 students in reading and math last summer.

Yi was able to help as a tutor and mentor, and one of the things we talked about is how kids need better phonemic awareness. This is called the "science of reading" which has been proven in paper after paper to be the most effective way to teach kids of all ability levels to read.

What it does

What Read Panda does is give tutors a fun and cute way of helping their kids read words and also find other words with the same sounds. By repeated exposure to these 1-4 character sequences, kids can make large improvements in their reading ability.

How we built it

We used GPT for a few different tasks, including figuring out a word's reading level, generating lists of words to practice, separating the word into phonemes, and most importantly finding new words that match those phoneme sequences. We used DALL-E to generate the images associated with the words.

The frontend was built with NextJS and Tailwind CSS.

Challenges we ran into

We had to learn a lot of new tools. This was Jiaqi's first introduction to GPT and prompt engineering. This was Yi's first time using the Dall-E API. We spent some time trying out the AWS text to speech and ultimately weren't able to get it working in time. Also, this was our first real introduction to the world of phoneme sequences and IPA symbolss.

Accomplishments that we're proud of

First and foremost, we had Cindy try it and she said "Kalen tried it but didn’t want to do a video. He struggled with Moon but when we went over the -oo- sound with with other words, by the 8th word, he was able to successfully read moon! 🤩" This is exactly what we want! To help kids read.

Secondly, even though Jiaqi isn't a professional designer she did an amazing job on the product design.

What we learned

Honestly, a ton. Even on things that we were already familiar with, the project helped us learn new tools and techniques, and obviously trying out new prompts and new k-shot examples was really interesting.

What's next for Read Panda

We're still working on getting each of the phoneme sequences to be pronounced independently. We also need to train the models to be a bit more accurate when it comes to identifying sequences and words with the same sounds.

It'd also be cool to integrate visemes at some point to teach the kids the mouth shapes when making the sounds. This has also been shown by research to help with reading skills.

Built With

• dall-e
• gpt
• nextjs
• vercel