AI agent integration is transforming GenAI development, but standardization remains the missing piece. 🧩
In this talk, Maks Operlejn, Senior ML Engineer, explores how Model Context Protocol (MCP) and modern frameworks are reshaping how we build production-grade AI systems.
You’ll learn about:
- Pydantic AI essentials – building type-safe, production-grade agents with structured outputs, dependency injection, and powerful evaluation tools,
- MCP fundamentals – Anthropic’s open standard for connecting AI models to external data sources and tools,
- Multi-agent system architecture – structuring data on demand with distributed, collaborative AI agents,
- 10 lessons learned – mistakes and breakthroughs from building scalable agent systems,
- Practical recommendations – best practices for creating secure, maintainable AI integrations.
If you’re building LLM-powered applications or exploring agentic AI workflows, this talk is your field guide to standardizing agent integration.
Timeline
00:00 Intro & agenda
02:45 Pydantic AI essentials
07:25 Agentic frameworks comarison
08:35 MCP essentials – what & how
13:20 Case study – structuring data on demand
16:41 Lessons learned: APIs & token budget
20:42 Lessons learned: tools, models & observability
25:01 Lessons learned: testing, guardrails & graphs
29:15 Lessons learned: security
30:35 Summary & key takeaways
Speaker
Maks Operlejn
Senior ML Engineer at deepsense.ai
