Lev Ronzhin
ENTERPRISE AI ENGINEER
«Don't know how to set up corporate AI and make it work for your business? I'm on my way!»
I design fault-tolerant corporate generative AI systems. My specialization is complex autonomous AI agents, building scalable in-house RAG architectures, and hardcore local inference of open-source models. I take on everything: from deep refactoring of corporate code to configuring queues and load balancers for LLMs under peak loads (300+ RPS).
Beyond server magic, I have a soft spot for automating routine tasks using custom AI apps. In my free time from vibe-coding, I test healthy recipes from the internet. Maybe I'll automate that one day too 😅
Core Competencies
GenAI, LLM & Inference
Tools: LangGraph, Agentic AI, LlamaIndex, LangChain.
Inference & LLMOps: Fine-tuning vLLM / llama.cpp for high loads. Load balancing and request proxying via LiteLLM, key rotation. Deep analytics and generation tracing via Langfuse.
Architecture: Model Context Protocol (MCP), Advanced / Graph RAG, Guardrails, HITL (Human-in-the-Loop).
Backend & Infrastructure
Development: Python, FastAPI, Asyncio. Creating CRUD services, refactoring corporate systems, systemic Code Review.
Databases & Queues: ClickHouse, PostgreSQL, Qdrant, Weaviate, ADQM, SAP HANA. Setting up message brokers for load balancing.
Deployment: Kubernetes, Docker, Apache Airflow, Prometheus, Grafana, GitLab CI/CD.
Work Experience
Severstal PJSC Senior AI / Tech Lead
11.2025 - Present | 20,000+ users | 300 RPS
Severstal PJSC Senior AI / Tech Lead
Latest achievements:
- Corporate in-house RAG: Designed and launched a custom in-house RAG (full CRUD service) from scratch with queues and load balancing, moving away from out-of-the-box solutions.
- LLM Proxy & Analytics: Implemented LiteLLM for request proxying, load balancing, and key rotation. Configured Langfuse for transparent AI quality and performance analytics.
- System Refactoring: Conducted massive refactoring of corporate AI system codebases and introduced strict Code Review standards. Upgraded local inference pipelines.
- Multi-Agent Workflow: Developed a LangGraph-based system for natural language SQL generation across heterogeneous DBs (ClickHouse, Postgres) utilizing a custom thesaurus (15k terms) via Qdrant.
Miramedix Data Scientist (NLP)
12.2024 - 06.2025
Miramedix Data Scientist (NLP)
- Developed NLP pipelines and agentic systems based on the LlamaIndex framework tailored for the medical domain.
- Implemented automated benchmarks to evaluate LLM response quality (Gemini, ChatGPT) using BERTScore, ROUGE, and BLEU metrics.
- Designed databases for embedding caching and generated synthetic datasets.
Institute of Digital Transformation of Medicine
03.2022 - 11.2025 | Tech Lead / Data Scientist
Institute of Digital Transformation of Medicine
- Architected and implemented a corporate DWH contour from scratch (PostgreSQL, MongoDB, Neo4j, Apache Airflow).
- Developed a CDSS prototype based on GraphRAG (Neo4j) and a medical RAG chatbot (FastAPI, Asyncio).
- Conducted fine-tuning of local LLMs (LoRA, QLoRA) with inference optimization via vLLM and llama.cpp.