Furkan Çelik

Built an internal knowledge assistant using the Gemini API and RAG architecture, semantically chunking institutional regulations to accelerate responses to student inquiries. Designed a classification system using sentence embeddings and similarity matching, labeling 1,000+ requests across 12 categories; the resulting LightGBM model achieved 79% accuracy, validated against manual review. Built an ELT pipeline and Power BI dashboards for monthly reporting of labeled complaints and requests.

Built a multilingual news clustering pipeline (embeddings + cosine similarity), outperforming legacy Solr. Benchmarked HDBSCAN/DBSCAN vs cosine-based methods. Built a FastAPI service for LLM comparison and a RAG-backed data entry system with function calling. Integrated PostgreSQL with Superset for client dashboards, replacing Power BI. Built custom Selenium/BeautifulSoup scrapers for hard-to-access sites.