Course : Autonomous Agents and Collective Intelligence, designing Distributed Ecosystems with LLMDesign and deploy scalable multi-agent systems integrating LLMs
Practical course - 3d
- 21h00 - Ref. LLM
|
Download in PDF format
Share this course by email
 | Identify the key concepts of intelligent agents and LLM-driven architectures. |
| Compare symbolic agents, LLM-powered and co-pilots to choose the right solution. |
| Implement a LangChain agent with tools, memory and reasoning chains. |
| Design a coordinated multi-agent system using roles, communication and workflows. |
| Apply ontologies and semantic reasoning to standardize and share knowledge. |
| Orchestrate and supervise multi-agent ecosystems, ensuring performance, security and scalability. |
Intended audience
Developers, software architects, AI engineers, researchers, R&D managers and any professional involved in complex projects in Python, LLM or distributed systems.
Prerequisites
Python language skills. Good knowledge of LLMs. Basic knowledge of LangChain. Good knowledge of software and agent-oriented architecture.
Course schedule
1 Foundations of intelligent agents and the role of LLMs
- Definition: agent, autonomy, environment, perception, action, objectives.
- Symbolic agent vs. LLM-powered. Difference between agents, chains and co-pilots.
- Difference between agents, chains and co-pilots.
- Overview of agent architectures (BDI, planners, prompt-based).
- Language model as a planning and reasoning engine.
- Limitations: hallucinations, coordination, computational cost.
2 LangChain agents and tool-based reasoning
- Architecture of a LangChain agent.
- Components: tools, memory, output parser, AgentExecutor.
- Examples of tools: calculation, search, files, APIs.
- Management of reasoning chains and the environment.
Hands-on work
Creation of a simple agent with tools. Agent that answers questions, uses a search tool, performs a calculation. Management of reasoning chains and environment.
3 Multi-agent theories and frameworks
- Agent types: specialized, hierarchical, competitive, collaborative.
- Different approaches to coordination: by task, by role, by message.
- Communication models: blackboard, publish/subscribe, direct dialog (JSON).
- Some frameworks: LangChain Multi-Agent, CrewAI, AutoGen, ChatDev, AutoGPT.
- Structuring complex workflows: delegation, roles, dependencies
4 Ontologies and semantic reasoning
- Standardizing shared knowledge.
- Integrating business ontologies into a multi-agent system.
- Reasoning from knowledge graphs (RDF, Neo4j).
- Long memory for inter-agent coordination.
Hands-on work
Conception d’un système à 3 agents spécialisés. Exemple : analyste de données, rédacteur de rapport, vérificateur juridique. Coordination par rôle et par objectif. Utilisation de mémoire longue.
5 Planning, supervision and safety
- Implementing chain-of-thought for LLM planning.
- The limits of LLMs: noise, instability, infinite loops.
- Control strategies: scoring, pruning, critical agents.
- Supervision and monitoring of interactions (logging, replay, auditability).
- Governance and security: sandboxing, compliance, critical agent or human-in-the-loop (HITL).
PARTICIPANTS
Developers, software architects, AI engineers, researchers, R&D managers and any professional involved in complex projects in Python, LLM or distributed systems.
PREREQUISITES
Python language skills. Good knowledge of LLMs. Basic knowledge of LangChain. Good knowledge of software and agent-oriented architecture.
TRAINER QUALIFICATIONS
The experts leading the training are specialists in the covered subjects. They have been approved by our instructional teams for both their professional knowledge and their teaching ability, for each course they teach. They have at least five to ten years of experience in their field and hold (or have held) decision-making positions in companies.
ASSESSMENT TERMS
The trainer evaluates each participant’s academic progress throughout the training using multiple choice, scenarios, hands-on work and more.
Participants also complete a placement test before and after the course to measure the skills they’ve developed.
TERMS AND DEADLINES
Registration must be completed 24 hours before the start of the training.
ACCESSIBILITY FOR PEOPLE WITH DISABILITIES
Do you need special accessibility accommodations? Contact Mrs. Fosse, Disability Manager, at psh-accueil@orsys.fr to review your request and its feasibility.
Developers, software architects, AI engineers, researchers, R&D managers and any professional involved in complex projects in Python, LLM or distributed systems.
PREREQUISITES
Python language skills. Good knowledge of LLMs. Basic knowledge of LangChain. Good knowledge of software and agent-oriented architecture.
TRAINER QUALIFICATIONS
The experts leading the training are specialists in the covered subjects. They have been approved by our instructional teams for both their professional knowledge and their teaching ability, for each course they teach. They have at least five to ten years of experience in their field and hold (or have held) decision-making positions in companies.
ASSESSMENT TERMS
The trainer evaluates each participant’s academic progress throughout the training using multiple choice, scenarios, hands-on work and more.
Participants also complete a placement test before and after the course to measure the skills they’ve developed.
TEACHING AIDS AND TECHNICAL RESOURCES
• The main teaching aids and instructional methods used in the training are audiovisual aids, documentation and course material, hands-on application exercises and corrected exercises for practical training courses, case studies and coverage of real cases for training seminars.
• At the end of each course or seminar, ORSYS provides participants with a course evaluation questionnaire that is analysed by our instructional teams.
• A check-in sheet for each half-day of attendance is provided at the end of the training, along with a course completion certificate if the trainee attended the entire session.
• The main teaching aids and instructional methods used in the training are audiovisual aids, documentation and course material, hands-on application exercises and corrected exercises for practical training courses, case studies and coverage of real cases for training seminars.
• At the end of each course or seminar, ORSYS provides participants with a course evaluation questionnaire that is analysed by our instructional teams.
• A check-in sheet for each half-day of attendance is provided at the end of the training, along with a course completion certificate if the trainee attended the entire session.
TERMS AND DEADLINES
Registration must be completed 24 hours before the start of the training.
ACCESSIBILITY FOR PEOPLE WITH DISABILITIES
Do you need special accessibility accommodations? Contact Mrs. Fosse, Disability Manager, at psh-accueil@orsys.fr to review your request and its feasibility.
Dates and locations
Select your location or opt for the remote class then choose your date.
Remote class
Dernières places
Date garantie en présentiel ou à distance
Session garantie