Brooke Duncan: Software Development with Agile Onboarding, Week 4

A proper start to the 4th of July: a hike with my mom and dog!

And all of a sudden, I have only one week left! Hello and welcome (back) everyone! My name is Brooke Duncan and I am doing a software development/engineering internship with Agile Onboarding here in Telluride, CO. As I am sure is the case for most of you, this past week has been primarily a week of barbecues, celebrations and spending time with family and friends. With that said, I did make some progress, mainly in the form of improvements, on my project. Our time and attention has been focused on fine-tuning the machine learning models that we are using, or, in this week’s case, changing them completely. On Monday we decided that the current ML model being used was not precise enough, and because there was no way for us to train it ourselves, we are in the process of replacing it with one that is more manipulatable. It will come as no surprise that this change is both time consuming and, at least on my behalf, difficult.

Some of the most important aspects of incorporating machine learning into prexisting software are ensuring that it: A) Is easy to use, B) Improves the user experience and/or backend operations, and C) Is accurate and consistent. Most of the issues we were running into revolved around the latter, the consistency of the information. I was finding that often times would receive different outputs when the inputs remained the same. It did not take long for Basit, my mentor, and I to determine that this issue was rooted directly in one of the preexisting LLM models we had implemented, and that this was not something we could change. Because of the holiday, I spent quite a bit of time independently researching and experimenting with a natural-language processing service called AWS Comprehend. This is a program that will allow us to take a large sample of our data and train the model to reconize certain information with much higher accuracy than the model we were using previously. The hope is that in the next week or so I (we) will be able to entirely replace what I was using before with this new program.  

Thanks for tuning in and I hope everyone has a great week!