Evgeniya Sukhodolskaya

Senior Developer Advocate at Qdrant, based in Munich. Works on Learned Sparse Retrieval — she authored Qdrant’s miniCOIL work and describes sparse neural retrieval as a personal research interest: “I really love sparse neural retrieval.” Co-hosts a search meetup in Munich and speaks at community conferences including MICES and Berlin Buzzwords.

She also authored Qdrant’s index-native Relevance Feedback work — a survey of the feedback literature followed by a mechanism that carries feedback into HNSW traversal, shipped in Qdrant 1.17.

Her recurring argument is accessibility: that sparse neural retrieval, model fine-tuning and relevance tuning are treated as ML-specialist territory when practitioners could be using them, and that tooling should close that gap. A second, related theme is that search engines shouldn’t be black boxes you build around — “search engines are here just to serve you a useful tooling and adapt to your needs and adapt to your users, be they humans or agents.”

Talks in this vault

• Evgeniya Sukhodolskaya - Fine-Tuning Sparse Neural Retrievers for E-Commerce — MICES 2026; walks from sparse vectors to a working SPLADE fine-tuning framework
• Evgeniya Sukhodolskaya - Relevance Feedback Inside the Search Engine — Berlin Buzzwords 2026; relevance feedback propagated into the HNSW hop-selection function

Articles in this vault

• Relevance Feedback in Informational Retrieval (2025) — survey of the feedback literature; the two-axis taxonomy and the research/production gap
• Relevance Feedback in Qdrant (2026) — the follow-through: formula, training procedure, BEIR results

Topics

• Learned Sparse Retrieval · Sparse Embeddings · SPLADE · miniCOIL
• Relevance Feedback · HNSW · Knowledge Distillation — the index-native feedback line of work
• E-commerce Search — intent-heavy retrieval as the motivating domain

Related

• Qdrant — employer
• Thierry Damiba — Qdrant colleague; she presented his fine-tuning experiments (Fine-Tuning Sparse Embeddings for E-Commerce Search)
• miniCOIL — her sparse retriever with BM25 fallback for out-of-vocabulary terms
• qdrant-relevance-feedback — the parameter-fitting package behind the relevance feedback query