Roman Shkarin
Senior Machine Learning Engineer @Remerge
Signup · Get unlimited contacts
WORK HISTORY
Senior Machine Learning Engineer @Remerge
Berlin, DE
Evolved a globally distributed real-time bidding system processing 480 billion daily requests (>5M QPS) and placing 4 billion bids daily, maintaining stable P95/P99 latency within a hard <50ms SLA- Optimized the hot-path of the main Go bidder service. Decoupled legacy architectural boundaries and built high-performance model inference interfaces, reducing Garbage Collection (GC) pressure while preserving the strict latency budget- Architected the bidding decision layer, upgrading CTR prediction to deep learning models (Self-Attention, DCN). Introduced probabilistic market landscape modeling to forecast price sensitivity curves, enabling surplus-maximizing bid optimization in a high-volatility, low-latency auction environment- Designed and built a unified Go/Python ML tooling ecosystem to standardize local-to-cloud workflows. Engineered a centralized CLI for pipeline orchestration and consistent artifact tracking (ensuring 100% reproducibility), alongside core libraries that eliminated boilerplate in model code. This directly reduced model time-to-production by ~3x- Drove the migration of the training infrastructure to Vertex AI. Re-architected core libraries to eliminate legacy bottlenecks and enable reliable processing of multi-terabyte datasets- Automated the end-to-end model promotion lifecycle. Engineered an artifact propagation system that reliably synchronizes ML artifacts across the globally distributed bidding infrastructure, ensuring strict version consistency and instant rollback capabilities- Built a deterministic A/B testing platform embedded directly in the bidding hot-path. Implemented orthogonal traffic partitioning and multi-dimensional targeting to safely evaluate concurrent ML models without violating real-time latency constraints.
EDUCATION
Karlsruhe Institute of Technology (KIT)
Doctor of Philosophy - PhD, Computer Science
Tomsk Polytechnic University
Master's degree, Computer Science
Tomsk Polytechnic University
Bachelor's degree, Computer Science
SKILLS
ABOUT ROMAN SHKARIN
Software Architect with 10+ years of experience combining distributed systems and research-driven ML to build ultra-low latency (5M QPS) platforms. Proven expertise in engineering real-time bidding systems, designing deep learning architectures, and orchestrating end-to-end MLOps lifecycles.
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.