AI & Data Science
I'm fascinated by how AI can solve real-world problems, so most of my projects start with "what if we could..." questions.
Turkish Social Media Analysis: I got curious about understanding sentiment in Turkish tweets - turns out it's trickier than English because of the language nuances. I fine-tuned VRLLab/TurkishBERTweet on small datasets and built models that can actually tell whether people are being genuine or sarcastic in their posts.
Smart Agriculture Solutions: Working with hyperspectral imaging for fruit ripeness detection was like solving a puzzle - you have hundreds of spectral bands but need to find the ones that actually matter. I used PCA and CARS to cut through the noise and was surprised how well classical Random Forest performed alongside AlexNet. I also built cow behavior tracking systems for predicting estrus, which sounds niche but is actually crucial for farmers.
Object Detection & Tracking: YOLOv8 became my go-to for real-time detection projects. There's something satisfying about watching objects move smoothly across video frames when you combine it with SORT algorithms for vehicle tracking.
Data Engineering & Analysis
Half the work in AI is getting your data right, and I've learned this the hard way through lots of trial and error. I build Python workflows that handle everything from scraping tweets with Selenium (while respecting rate limits) to cleaning messy datasets and preparing them for analysis.
I spend a lot of time with Pandas and NumPy doing exploratory data analysis - those moments when you finally spot the pattern in your data make all the debugging worth it. My approach is straightforward: get the data, understand what you're actually looking at, clean it properly, and document everything so you can reproduce your results later.
Machine Learning Projects
I'm more interested in creating working prototypes that people can understand and use rather than claiming I've built the next revolutionary AI system. Every project I work on, I ask myself: "Does this actually solve a problem? Can someone else understand how it works?"
Whether I'm fine-tuning BERT models for sentiment analysis or implementing computer vision systems for agriculture, I focus on building things that are practical and honest about their limitations. I'd rather have a simple model that works consistently than a complex one that fails unexpectedly. The field moves fast, but I think the fundamentals matter: clean data, understandable code, and solutions that address real problems.