Fine-tuned MCQ Qwen-Instruct
Optimized a Qwen2.5-0.5B model for machine learning reasoning via Supervised Fine-Tuning (SFT). Developed a high-signal dataset of 985 samples using synthetic and curated sources, resulting in a 1st place finish (57.6% accuracy) in a competitive cohort of 417 students.
Speech Accent Detection
Engineered a pipeline to transform raw audio into STFT Spectrograms for "accent image" classification. Using a custom 400k-parameter ResNet, I stabilized training to achieve 62% accuracy in identifying L1 language families—priming ASR systems for more inclusive recognition.
UEFA Euro's Match Predictor
Developed an end-to-end ML pipeline to forecast UEFA Euro 2024 outcomes. By programmatically scraping 60+ years of match data via BeautifulSoup, I trained a Random Forest Classifier that successfully predicted Spain as the tournament champion.