Aleksandr V. Petrov

Applied ML Researcher | Search, Recommendation & Generative AI

Industry focus: Production LLM systems, retrieval and reranking, robust evaluation, large-scale personalisation

Highlights: Spotify Home and Search | +3% search CTR at Tripadvisor | RecSys 2023 Best Paper | SIGIR, RecSys, WSDM

firexel@gmail.com+44 7455 908423Glasgow, UK (remote)British citizen; no visa required

Core expertise

September 2025 - present | Glasgow, UK (remote)

Spotify Home: Contributed to a team building hypothesis-driven shelf generation; led evaluation work across the LLM planning, generative retrieval, candidate alignment, and offline serving pipeline.
Personalised Search: Contributed to an LLM cross-encoder combining semantic, behavioural, and user-context signals; developed the initial training recipe. The resulting system improved offline ranking by 13.3% and live search success by approximately 2%.

2021 - July 2025

Full-time career break: research on large-scale recommender systems with language models; viva with distinction and no corrections.

Amazon, 2022 & 2024: fixed-term Applied Science engagements during the PhD; built an identity-free advertising training pipeline and user-modelling methods.
Viator, March - August 2025: bridge back to full-time industry; deployed semantic search (+3% CTR), developed LLM-based position-bias estimation, and contributed to writing the GLoSS paper.

2017 - 2021 | Edinburgh, UK

Built PeerSets, an ML competitor-identification system integrated into 20 advertising products; coordinated approximately 10 partner teams.
Led technical work across recommendation, search ranking, personalisation, and advertising.

1 / 2

Aleksandr V. PetrovApplied ML Researcher

2013 - 2017

Built a white-label recommender platform for video, music, and news companies.

Data-Centric Alliance - Head of R&D 2013 - 2016

Tinkoff Digital - Data Platform Team Leader 2012 - 2013

Mail.Ru Group - Personalisation Researcher 2012

Yandex - Software Development Engineer 2009 - 2012

2021 - July 2025

Large-scale recommender systems with language models. Supervisors: Prof. Craig Macdonald and Prof. Iadh Ounis.

2006 - 2011

Master's-equivalent degree; thesis awarded an excellent grade.

Co-organiser: Unified Search and Recommendation Workshop and FRAME, RecSys 2026.
Co-instructor: Transformer-based Sequential Recommender Systems tutorial, RecSys 2026.
Programme committee: RecSys, WWW, ECIR. Reviewer: TOIS, TORS, TKDD.

Hypothesis-Driven Shelf Generation for Personalised Recommendation. RecSys 2026 Industry Track. paper
Robust Fusion of Semantic and Behavioural Signals for LLM Reranking in Personalised Search. RecSys 2026 (USRW workshop).
Do LLM-judges Align with Human Relevance in Recommender Evaluation? RecSys 2026 (USRW workshop). paper
Do Sequential Recommendation Benchmarks Really Require Higher-Order Sequence Modelling? RecSys 2026.
LLMs for Estimating Positional Bias in Logged Interaction Data. RecSys 2025 (CONSEQUENCES workshop). paper
Generative Sequential Recommendation with GPTRec. SIGIR 2023 (GenIR workshop). paper Most cited publication.
Efficient Recommendation with Millions of Items by Dynamic Pruning of Sub-Item Embeddings. SIGIR 2025. paper
RecJPQ: Training Large-Catalogue Sequential Recommenders. WSDM 2024. paper
gSASRec: Reducing Overconfidence in Sequential Recommendation. RecSys 2023. paper Best Paper Award.
RSS: Effective and Efficient Training for Sequential Recommendation using Recency Sampling. ACM TORS 2023. paper
A Systematic Review and Replicability Study of BERT4Rec for Sequential Recommendation. RecSys 2022. paper

2 / 2