Computer Engineering & Mathematics @ University of Warwick. I’m interested in applied ML research (deep learning, probabilistic modelling) and currently work on ANN / vector search.
I’ve also built production-facing data and ML pipelines in startups (Python, SQL, AWS, Docker).
Links
- Email: roman.bikbulatov@warwick.ac.uk
- GitHub: github.com/zodackwand
- LinkedIn: linkedin.com/in/romanbikbulatov
Research
Architecture-Aware Vector Search (C++ ANN indexing)
University of Warwick - supervised by Prof. Hakan Ferhatosmanoglu - Jul 2025–present
I focus on filter-aware graph traversal for approximate nearest neighbor (ANN) search: how to handle selective constraints efficiently while maintaining strong retrieval quality.
What I do:
- Design and test traversal strategies for selective / filtered queries.
- Run controlled experiments and track relevant metrics.
- Workloads: SIFT1M and larger synthetic datasets on a 500GB RAM HPC cluster.
(If useful for reviewers: I can share a short technical note / results snapshot.)
Selected engineering work
Extruct AI — Software Engineering Intern - Jun 2025–Sep 2025
- Built a Python + SQL attribution pipeline for a B2B data product.
- Implemented backend components for a people search microservice (Python, PostgreSQL).
ClearPic — ML/NLP Intern - Jun 2024–Sep 2024
- Built an NLP preprocessing pipeline for a 10M+ record dataset (Python; HuggingFace, spaCy, NLTK).
- Containerised and deployed the pipeline as a repeatable batch job (Docker, AWS ECS).
Research interests
- Retrieval & representation learning: ANN, vector search
- Efficient ML systems: latency, memory locality, tail behaviour, reproducibility
- Probabilistic / stochastic modelling for ML and decision-making under uncertainty
- Optimisation + scientific computing / HPC for data-intensive workloads