Junior ML-инженер

• Developed a two-stage semantic search and ranking system (bi-encoder + cross-encoder) for corporate knowledge base, improving Recall@20 by 15% compared to BM25 baseline and reducing the share of "empty" search sessions. • Developed NER (Named Entity Recognition) for structuring and annotating raw documents: extracted key attributes(dates, numbers, titles, organizations), improving search accuracy by metadata and simplifying document navigation. • Solved the long document challenge (2000+ tokens) through dynamic chunking and score aggregation, maintaining consistent quality on lengthy documentation. • Designed and deployed ML services (FastAPI, Docker) with logging and monitoring, enabling continuous quality assessment and rapid iteration based on user feedback.