It's crazy to think that 10% of this challenge has already passed. To be honest, it feels like I haven’t accomplished as much as I hoped.
Looking at the statistics, I’ve spent only 23 hours studying AI, which is 7 hours below my target for this stage.
One key observation is that I spent most of this time on projects:
Unfortunately, many of these projects were more about web development than AI. Nevertheless, here’s a recap of the projects I’ve worked on so far:
Projects
1. REST API using FastAPI to Run Langchain Code in Production
This project is still in progress, and it's for work. While it involves AI in a sense, I’m mostly calling OpenAI endpoints, which feels a bit underwhelming. However, I am learning a lot about the challenges of deploying AI, like handling large data transfers and securing the server to prevent misuse of the service.
My biggest issue, though, is the Python programming language. I absolutely dislike working with Python. Every time I use it, I miss the type safety of languages like Java and TypeScript. I’ve considered switching to Mojo, but since this project is for work, it needs to be easily maintainable. Mojo is still in its early stages and lacks a robust web framework. So, for now, I’ll have to put up with Python. On the bright side, this gives me a chance to understand why Python is so beloved and widely used in AI.
2. News Article Bias Detector
I followed a mini-course on ZTM that showed how to use Hugging Face to train a text classification model. Using what I learned, I scraped news articles from various websites and tried detecting bias on specific topics. I then used Hugging Face's transformers and datasets libraries to train a DistilBERT model.
It was a small project, but I was proud of the results. After just 10 epochs, the model achieved 96% accuracy.
3. Streamlit Jira Statistics Dashboard
For work, I was asked to gather statistics on our sprint performance. I did this by connecting to the Jira API and using Streamlit to create a beautiful dashboard. The dashboard provides insights like average time spent per story point and the overhead per developer and for the overall team.
One fun feature I added was the "Most Helpful Developer" metric, which checks who logged the most time working on tasks assigned to other team members. It was a neat way to recognize cross-team collaboration.
Conclusion
Overall, I notice that I’m still doing more software engineering tasks than actual AI work. I understand that AI is a subset of computer science, and these skills are valuable, but I feel like I need to focus more on isolated AI practice to make the most progress where I have the least experience.
