About Me
👋 Hi, I'm Slava Dubrov — an seasoned AI/ML engineer with 10+ years of shipping production ML and AI systems and leading ML / AI teams. I currently work on LLM deployment, fine-tuning, and agent optimization on the Agent Execution team at HubSpot. Before that, I spent years building the retrieval and memory infrastructure that agents rely on — so when they hallucinate, I take it personally.
Why You Should Read This Blog
I write about what actually works in production AI systems. My perspective comes from:
- Working on Agent Execution at HubSpot — LLM fine-tuning, inference optimization, agent evaluation, and safety guardrails in production
- Worked on the Embedding Hub and Context Layer for AI Agents — the retrieval, grounding, and memory infrastructure powering autonomous agents across all HubSpot products
- Saving companies millions — $4M+ yearly at Wayfair alone through fraud/scam detection and embedding systems I built and led
- Speaking at industry events — presented "Engineering the Agentic Stack" at the World Agentic AI Summit in Berlin (2026)
- Publishing research — PhD in AI diagnostics, peer-reviewed publications, and patents in intelligent systems
- Hands-on with the full stack — from data pipelines and model training to evaluation, deployment, and safety guardrails
- Cloud ML architecture expertise — designing and deploying AI systems across AWS and GCP as batch, streaming, and real-time services
- Open source contributions — code, tutorials, and guides for practical AI engineering
Speaking
- "Engineering the Agentic Stack" — World Agentic AI Summit, Berlin (2026). Production architecture for agentic AI systems covering Cognitive Engine, Cortex (memory architecture), and Schema-Guided Reasoning.
What I Write About
My posts focus on practical AI engineering:
- Agent Architecture — Tool Ergonomics & ACI, Memory for AI Agents, Cognitive Engine
- Context & Retrieval — Context Engineering for agentic systems, RAG patterns
- LLM Development — Fine-tuning guide, Schema-Guided Reasoning, LoRAX Playbook
- Developer Tooling — Python setup, uv, MCP servers
Tech Radar
LLM Serving & Fine-tuning: vLLM · LoRAX · LoRA/QLoRA · VLMs · SGR/SO
AI Agents: LangGraph · Claude · Google ADK · CrewAI · LlamaIndex · SmolAgents
Agent Safety & Evaluation: Guardrails · Automated evals · LLM-as-a-judge · Observability
Vector & Retrieval: Qdrant · Faiss · Semantic search · Hybrid retrieval · Reranking · Context compression
Tool & Workflow Integration: MCP (Model Context Protocol) · A2A · FastMCP · n8n
MLOps: AWS (2 certs) · GCP/Vertex AI · Kubernetes · Kubeflow · Airflow · Ray · MLFlow
Core: Python · SQL · Scala · Java · Rust · PyTorch · FastAPI · Spark · Polars
Let's Connect
Let's connect — especially if you're tackling hard ML problems at scale, or if you just need someone to confirm that yes, your pipeline is supposed to be that complicated.