This week was all about diving into the AI side of JAICE and figuring out what models we will be using to power our email parsing and classification features. We spent most of our time researching and breaking things down into three main models that will each handle a different part of the process.
The first is a binary model that determines whether an email is job-related or not. This will help us filter out everything that is not relevant, so JAICE only focuses on messages that matter to the job search.
The second model, which I focused on this week, is a classification model. Its job will be to sort each email into the correct category, like “Application,” “Interview,” “Offer,” or “Accepted/Rejected.” For this one, we are exploring a zero-shot classification approach, which is really exciting because it does not require us to manually train the model on every possible label. Instead, it can understand new categories based on context, which makes it more flexible and scalable.
The third model will handle Named Entity Recognition (NER) to extract important details such as the company name, job title, and other information we will want to display for each application to the user.
Outside of the AI research, I also made a small but useful UI update. Each job card now includes a “View Email” link. This lets users open the email directly if they want to see more details, instead of having to log in and search for it manually.
Next week, we are planning to start testing some of these models to see which ones integrate best with our current setup.
