**Build Log: Zero to AI Project – Day 14**
Hey team!
What a whirlwind of a couple of days it’s been! We’re definitely picking up steam on our Zero to AI project, and I can’t wait to share all the juicy bits from the last 48 hours.
First off, our data collection phase is finally complete! We successfully gathered a solid dataset that has both variety and quality—gotta give a shoutout to the team for really nailing down the scraping and cleaning process. We encountered some hiccups with incomplete data and inconsistent formatting at first, but after a few rounds of tweaking our scrapers and leveraging some nifty data augmentation techniques, we whipped it into shape. Who knew data wrangling could be such a wild ride?
On the algorithm front, we’ve made solid headway with our model! After running a few initial tests with different architectures, we decided to pivot toward a hybrid approach, pulling in elements from both CNNs and Transformers. It sounds complex, but so far, it’s really starting to show promising results. We’ve optimized our training regimen too, thanks to some trial and error with learning rates and batch sizes. This led to a 15% increase in accuracy, which is a huge win for us!
However, we did hit a wall with overfitting during our last training round, which was a bit frustrating. But don’t worry, we researched dropout layers and implemented them—fingers crossed that this will keep the model generalizing better.
As for next steps, we’re shifting our focus to fine-tuning the model and preparing for our first round of user testing within the next week. We’ll also start prepping some documentation and a demo to showcase what we’ve built so far. I’m excited to see how our users interact with the model and what insights we can gather from their feedback.
That’s the scoop! It’s been a rollercoaster, but I feel like we’re building something really cool and impactful. Time to grab some coffee and dive back in—let’s keep the momentum going!
Catch you later!