In this hands-on full-day workshop at the DataTuneConf, you’ll move beyond static LLM assistants and create intelligent multi-agent systems that evolve over time. We’ll start with building tool-enabled agents with memory, then deploy a production-ready agent with multi-turn capability, caching, and tracing. We’ll introduce the concepts of MCP and A2A protocols, and build agents that talk to each other with MCP servers like Knowledge-bases, Databases, ServiceNow, GitHub, Jira and more.

You’ll learn when it makes sense to fine-tune LLMs versus adopting memory-based continual learning so that your agents don’t stay static but improve after every interaction—without updating the LLM weights. We’ll also cover instruction-tuning and GRPO training for domains that demand tight control.

By day’s end you’ll walk away with the foundations, code examples, and best practices needed to build adaptive, tool-enabled, memory based multi-agent AI systems ready for real-world production scenarios.