Towards Agentic Intelligence: Architectures for Multi-Agent AI Systems
Building on the tremendous response to Dhanashree Lele’s ACM talk on Multi-Agent Architectures for Enterprise AI, this 2-day, research-caliber, hands-on workshop is designed to advance the state of practice in Agentic AI system design, evaluation, and optimization.
This workshop will guide participants through theory-to-deployment workflows for constructing next-generation multi-agent frameworks, benchmarking agentic behaviors, and applying compute-efficient orchestration strategies. The curriculum draws heavily from recent breakthroughs presented at NeurIPS, ICLR, and KDD, grounding hands-on engineering in rigorous scientific principles and reproducible experimentation.
By bridging academic research and production-grade engineering, this workshop is ideal for applied researchers, industry practitioners, graduate students, and technical leaders seeking to design reliable, interpretable, and high-performance LLM-based agentic systems.
Learning Format and Structure
This intensive two-day workshop follows a progressive “build-as-you-learn” methodology. Each module introduces core research concepts followed by guided implementation in Jupyter/Google Colab, enabling participants to translate theory directly into working systems.
📅 Day 1 — Architecting Multi-Agent Systems
4 hands-on labs focused on:
-
- Building multi-agentic systems/use-cases from first principles
- Exploring agent tools, MCP operability, and orchestration strategies
- Converting agentic prototypes into robust, production-ready cognitive pipelines
- Implementing coordination, planning, and tool-use protocols across agents
📅 Day 2 — Deep Dive: Research Frontiers and Reproduction
- Analytical walkthroughs of seminal and frontier papers in Agentic AI from NeurIPS, ICLR, and KDD
- Structured methodology for paper-to-prototype translation: reproducing cutting-edge research through practical labs
- Discussions on evaluation benchmarks, alignment frameworks, and emergent behavior analysis
- Roadmapping techniques for embedding research-grade systems into real-world enterprise use cases
Key Outcomes
Theoretical Foundations:Understand the mathematical and algorithmic underpinnings of multi-agent LLM architectures, orchestration, and alignment.
Hands-On Mastery:Gain practical experience in building agentic systems from scratch, configuring MCP operability, and scaling prototypes into production-grade deployments.
Evaluation & Governance:Learn to design and apply alignment and evaluation frameworks to ensure robustness, interpretability, and responsible deployment of multi-agent systems.
Practical Assets:Walk away with fully functional notebooks, baseline reference architectures, curated reading lists, and reproducible workflows to accelerate implementation in your own organization.