This week, I focused on integrating the classification model into JAICE, which was a big step for our AI features. The model now takes job-related emails and automatically labels them based on what stage they are in, like Applied, Interview, Offer, Accepted, or Rejected. Seeing it start to categorize messages on its own was really exciting after all the research we did last week.
In addition to labeling, the model also stores a confidence score for each prediction. This will come in handy later when we start adding user feedback. If a score falls below our threshold, JAICE will flag it for review and ask the user to double-check the classification accuracy. That way, we can make sure users see the right status for their applications instead.
Right now, the system is only storing these confidence scores in the database, which is a good first step. The model itself seems pretty accurate so far, though we still need to do more testing with larger datasets to be sure. One thing we have noticed is that it takes a little while to process and classify each email. After discussing it as a team, we agreed that we would rather have it be slower but more accurate than fast and wrong.
Next week, we will continue fine-tuning the model and testing different thresholds to strike the right balance between speed and accuracy as we move closer to integrating it fully with the rest of JAICE.
