Open to opportunities

Mikhail Kalinin

@mikhailkalinin

Production AI Engineer specializing in multi-agent LLM systems, cost-efficient deployments, and measurable automation impact.

Sweden

Message

What I'm looking for

I seek hands-on roles building scalable, production-grade LLM systems and multi-agent automation with strong observability, guardrails, and measurable business impact.

I am a Production AI Engineer with 6+ years delivering multi-agent LLM systems and end-to-end solutions from requirements to production monitoring.

I build autonomous agents, Agentic RAG, and production RAG systems with strong guardrails and observability, achieving significant operational savings and performance gains.

My work includes fine-tuning and alignment (QLoRA, TRL, DPO/GRPO, Unsloth), context engineering, model quantization and rigorous LLM evaluation to reduce hallucinations and improve reliability.

I focus on scalable, cost-efficient deployments (vLLM, Docker, Kubernetes, AWS/GCP) that translate cutting-edge research into measurable business outcomes such as OPEX reduction, workload automation, faster analytics, and multi-million dollar savings.

Experience

Work history, roles, and key accomplishments

Current

Production AI Engineer

Current

Tele2 Sweden

Aug 2023 - Present (2 years 7 months)

Built multi-agent autonomous AI systems that reduced OPEX 54%, automated 67% of processes, and cut HR time-to-hire by 69%; deployed RAG and vLLM solutions delivering 94% accuracy and 10× cost savings.

RAG VLLM MCP QLoRA LangGraph Docker Qdrant Model Monitoring

Machine Learning Engineer

Deutsche Bank

Mar 2022 - Jul 2023 (1 year 4 months)

Developed production Transformer-based customer behavior and recommendation models improving prediction accuracy +18% and conversion +10%; built GDPR-compliant fraud detection saving $1.2M annually.

Transformers Recommendation Systems Fraud Detection Docker Airflow CI CD Model Monitoring Spark Snowflake

NLP Engineer

AdPulse

Feb 2020 - Mar 2022 (2 years 1 month)

Built NLP pipelines for brand safety and sentiment analysis (BERT/DistilBERT) yielding F1=86% and increasing advertiser conversions +12%; improved audience segmentation driving significant ad revenue.

BERT DistilBERT NLP Sentiment Analysis Embeddings K Means A B Testing

Education

Degrees, certifications, and relevant coursework

Lomonosov Moscow State University

Bachelor of Science, Computational Mathematics and Cybernetics

2019 - 2023

Grade: 4.5

Bachelor of Science in Computational Mathematics and Cybernetics with specialization in Applied Mathematics and Computer Science, focused on computational methods and software for data analysis and modeling.