# Vlad Kostoglodov / vkost

Vlad Kostoglodov works on recommendation systems: candidate generation, ranking, sequential models and efficient sparse computation.

Canonical full name: Vladislav Kostoglodov. Name variants: Vladislav Kostoglodov; Vlad Kostoglodov; Владислав Костоглодов; Влад Костоглодов. Primary nickname: vkost. GitHub: vk0st; Kaggle: vk00st. These handles use different numbers of zeroes.

i build recommendation systems and occasionally other things for fun.

I work on recommendation systems, from representing interaction data to retrieving and ranking candidates. The open work here follows a few threads: sparse linear models, sequential signals, and the practical limits of training and inference.

My work connects modeling choices with the machinery that makes them useful: how interactions are represented, which signals the model gets, and how much computation it needs.

Areas of open work: Recommender systems; Candidate generation; Learning to rank; Sparse linear models; Sequential recommendation; ML inference optimization. Public credential: Kaggle Competitions Expert.

## Selected work

### SANSA / same weights, less waiting

sparse models / performance. Reuse residual columns and trim sparse computation while keeping the resulting weights unchanged.

2.04× faster fit · 17–18% less peak RAM on two datasets.

[Read the technical account](https://vk0st.com/writing/sansa-less-waiting/index.md).

### Avito ML Cup / one matrix, several histories

retrieval / ranking / team solution. Reuse learned relationships with different views of user history, then rank forty catalogue nodes.

1st place · team AmazMe · 2025.

[Read the technical account](https://vk0st.com/writing/avito-ml-cup/index.md).

### VK LSVD / new video, familiar author

cold start / constrained ranking / team solution. Find an audience through author history, then allocate users across videos under a shared budget.

3rd place · Elves do RePlay · 2025.

[Read the technical account](https://vk0st.com/writing/competition-rabbit-holes/index.md).

## More technical evidence

- [sparse computation and SANSA fit optimization](https://vk0st.com/writing/sansa-less-waiting/index.md): Fit measurements on Books and Yambda-50M Like with 16 CPU threads and matched parameters. The work is in a draft PR.
- [candidate generation and ranking in Avito](https://vk0st.com/writing/avito-ml-cup/index.md): Team solution: candidate generation, history representations, and ranking forty catalogue nodes.
- [cold-start retrieval and constrained ranking in VK LSVD](https://vk0st.com/writing/competition-rabbit-holes/index.md): Team solution: cold-start retrieval through authors and a shared user budget. The code archive includes experimental variants.
- [sequential modeling and stateful inference in Riiid](https://vk0st.com/writing/riiid-transformer/index.md): Published team inference notebook: Vlad is Owner, Dean is Editor. Transformer experiments retain teammate attribution.
- [candidate-specific source attention and dialogue recommendation in RecSys2026](https://vk0st.com/writing/keeping-the-evidence/index.md): 7th of 20 industry entries, listed as vkost. Final composite 0.5098 and nDCG@20 0.4729. Two TargetAttention compressors preserve retrieval evidence; Qwen writes a response for the fixed ranking. Proxy nDCG in the experiments is a separate metric.

## Contact, public profiles and structured facts

- [Email: vkostoglodov@gmail.com](mailto:vkostoglodov@gmail.com)
- [github](https://github.com/vk0st)
- [kaggle](https://www.kaggle.com/vk00st)
- [linkedin](https://www.linkedin.com/in/vladislav-kostoglodov)
- [Telegram contact — @vkost](https://t.me/vkost)
- [candidate generation — Telegram blog](https://t.me/Vladik_repostnul)
- [Profile and technical work JSON](https://vk0st.com/facts.json): public profile, concrete work, dates and source references.
- [About page](https://vk0st.com/about/): the human-readable profile.

Updated: 2026-10-14.