T-Technologies brought open-source Scikit-Rank for neural ranking...

T-Technologies brought open-source Scikit-Rank for neural ranking to ACM RecSys 2026

T-Technologies teams (T-Bank ecosystem) presented an open-source library Scikit-Rank with a scikit-learn-compatible interface to neural rankers DCN v2, FinalNet, FinalMLP, Destine, and TabM, as well as the Perseus framework for cross-service personalization and the ScaL³AE method for scaling LLM-enriched linear autoencoders to catalogs of millions of objects, at the ACM RecSys 2026 conference.

Compiled by Sergey KostenchukPublished 2026-09-30Updated 2026-09-30

2026-09-30 Research HuggingFace

🤖 Open-source Scikit-Rank shown at ACM RecSys 2026

T-Technologies (T-Bank ecosystem) presented three works. The main one is open-source Scikit-Rank: rankers DCN v2, FinalNet, FinalMLP, Destine, and TabM in a sklearn fit/predict interface, more than 20 loss functions — a replacement for CatBoost/LightGBM with almost no code changes. Alongside it are Perseus and ScaL³AE.

🌍 According to the authors of Perseus, a single model on events from all company services raised recommendation quality from 16 to 39% on T-ECD (metric not disclosed). This is a trend for multi-product ecosystems; cross-service data processing will require legal review.

👤 Scikit-Rank is installed via pip install scikit-rank and works on CPU and CUDA (PyTorch, Hugging Face Accelerate), the repository is open. For RecSys practitioners — a map of industrial solutions from a major Russian player.

Source 1: https://habr.com/ru/companies/tbank/articles/1088224/

Source 2: https://dl.acm.org/doi/10.1145/3773078.3831815

Sources