Company
SberDevices
5 cases from SberDevices (General, Search engine), each digested from a public talk or engineering post and linked to the original. 1 report results in production.
Tasks: Build an assistant over internal docs, Rank search results, Learn user and item embeddings, Evaluate models. Methods: Data augmentation, Triplet loss, Crowdsourced labeling, RAG, Cross-encoder reranking, Tool / function calling and 17 more. Tools: Toloka, GigaChat, OpenSearch, XLM-RoBERTa, CatBoost, FAISS and 12 more.
GigaSearch, or a Search System on GigaChat![]() | ![]() | search-engine | 2024-05 |
RLHF for retrieval based NLP system![]() | ![]() | general | 2023-05 |
Majority vote for CV annotations: improve your data labeling pipeline![]() | ![]() | general | 2023-05 |
Modern training methods for retrieval-based dialogue models![]() | ![]() | general | 2022-06 |
Building text summarization for Russian-language texts![]() | ![]() | general | 2022-06 |
