Created by: Willa Kong, Stephen Lin, Mark Branton, Azin Taqizadeh, with Team ID 0B9B6

SupremeUofT - Spellerly

Autocorrect for handwriting on a tablet

These days, many students have been converting from traditional pen and paper writing to tablet writing. However, in this digital age, the benefits of autocorrect have not yet caught up to this innovative note-taking. So, using the power of OCR and NLP, we have created Spellerly, the tool to autocorrect handwriting.

We learned about machine learning as we had a chance to work with NLP and training models for it. In addition, we were able to learned more about messaging services which are similar to well-known services such as Apache Kafka and IBM MQ. We also worked with hardware and were able to push the bounds of the capabilities of the Raspberry Pi using our current knowledge.

So, this was built using a Raspberry Pi and display to simulate a tablet a student may use. Then, using the message broker Solace, the Raspberry Pi would send the note that was processed with OCR to our Python service that would then take the given text and determine if there are any spelling mistakes that need to be correct. In particular, it provides a suggested replacement using a custom-trained, TextBlob model that would find the closest words that matches the intended meaning.

The challenges we faced included viability of our original ideas, hardware issues, as well as connection issues with the message broker. In the beginning, we had a much larger scope for our original idea but as time progressed, we realized we had to narrow the scope as it took too long to debug some of our problems. In terms of the hardware, we had a tough time connecting to our LCD screen but in the end, we borrowed another group's screen that we knew for sure worked. Additionally, the eduroam network was not connecting to our Raspberry Pi and, as a result, we had to go and purchase an ethernet cable and use hotspot for its WiFi. Finally, as Solace is a budding message broker, there are not a lot of problems that have been explored and thus the problems we faced took some time to identify and solve.

Built With

Share this project: