At Idea Lab, FSB, I fine-tuned a conversational LLM to answer preparatory-class student questions, grounding responses with retrieval-augmented generation and prompt design. I also collected, cleaned, and structured its training corpus through web scraping.
In my Parkinson’s disease detection project, I reduced 22 acoustic features to six and benchmarked five classifiers, reaching 90% accuracy and 97% recall. I’ve also built a reinforcement learning agent evaluated against a greedy baseline and contributed data analysis to academic projects on power-cut equity and tourism.
