Artem Sidorov

Senior AI / NLP Learning Engineer

XSOLLA

Text-to-SQL: Built a secure agent over about 2,000 table descriptions with schema retrieval, reranking, SQLGlot AST validation, EXPLAIN, scoped read-only roles, and bounded repair. Achieved 82% SQL Success@1, 94% top-5 schema recall, and 85% analyst-free completion after repair, manual requests fell from about 400 to 240 per week. • Customer-support RAG: Built a multilingual, on-premises RAG platform over 5,200 documents | 82,000 chunks using Qdrant, Elasticsearch BM25, RRF, a domain-tuned multilingual-e5 embedder, and bge reranking, achieved 93% Recall@10 and 91% nDCG@10 on held-out real supp

Machine Learning Engineer

inDrive (formerly inDriver)

Internal RAG: Led ML for a bilingual knowledge assistant used by about 1,200 employees across 10,000 documents. Hybrid retrieval and reranking raised Recall@10 from 82% to 91% and nDCG@10 by 7 points, query rewriting and citation validation cut unsupported answers from 8% to 2% and search time from about 90 to 10-15 seconds. • Ticket classification: Built multilingual support routing from 500,000 historical tickets, including a 15-class taxonomy and a 15,000-example re-annotation workflow. Achieved 0.89 macro-F1 | 0.92 weighted-F1, automated 85% of routing, and reduced errors from 18% to 9%; l

Bachelor of Computer Science

Moscow Institute of Physics and Technology (MIPT)

Master of Data Science

Nazarbayev University