Skip to content
DSML.KZ
RN

Ruslan Nagimov

@kapalaq

ML Engineer with strong System Architect mind

I am a Data Scientist at Beeline Kazakhstan. I have been working with classical Data Science, Recommendation Systems and LLM-based solutions for various tasks

Open to

Project collaboration

Career

Experience: 2 years

Experience

Data Scientist (ML Engineering Specialized)

Beeline Kazakhstan (Veon Ltd.)

Telecommunications operator in Kazakhstan.

Aug 2025 — Present

Built production ML, ranking, search, and LLM serving systems; reported improvements in resolution, ranking metrics, chat coverage, and latency.

• Developed and deployed a production ML inference pipeline handling 100 QPS throughput across distributed microservices (Kubernetes, Docker, Apache Spark, Milvus vector DB) for enterprise CRM system chat bot. • Trained a LambdaMART reranker on session-outcome labels using BM25, semantic similarity, and historical resolution-rate features, improving article resolution rate by 20%, MRR@5 by 19%, and Precision@1 by 25% while maintaining 150 ms serving latency; achieved a 32% chat coverage increase for AI-assisted operators during A/B testing.

Data Scientist Intern

Beeline Kazakhstan (Veon Ltd.)

Telecommunications operator in Kazakhstan.

Feb 2025 — Aug 2025

Developed a Markov Chain next-song recommendation MVP that achieved NDCG@5 of 0.79.

• Researched and developed an MVP of a Markov Chain model for next-song prediction, achieving NDCG@5 = 0.79 in seamless playlist recommendations.

Education

B.Sc., Computer Science

Nazarbayev University

2023 — 2027

Selected projects

Meta Llama Accelerator

Built an end-to-end agentic system with a feedback-driven SFT pipeline that converted human ratings and corrections into training targets via a custom trajectory abstraction (ART), fine-tuned with SFTTrainer (TRL) and LoRA; optimized training with a 6× speedup through Unsloth. Implemented vector retrieval (Milvus) for semantic search, achieving 88% answer relevance in labeled tests.

HackNU/26 Hackathon

Architected event-driven infrastructure based on Redis Streams, Pub/Sub, and WebSockets for robust, scalable asynchronous ETL workflows.

Achievements

Meta Llama Accelerator — 2nd Place

Meta (in collaboration with BAITC Ltd.)

HackNU/26 Hackathon — 2nd Place

KazakhTemirZholy in collab. with NU ACM SC

Skills

Classic MLRAG ArchitectureRecommendationRankingLLMsStatisticsA/B testingPandasNumPyscikit-learnKubernetesDockerFastMCPApache SparkGit/GitHub ActionsPostgreSQLMilvusRedisMinIOPython