Designing AI agents — bootcamp

Objectives

• Call a language model through its API with structured outputs, error handling and control of the cost per request
• Build a document search pipeline (ingestion, chunking, embedding, retrieval, source citation) on a supplied corpus and measure its quality on a set of test questions
• Equip the agent with tools (functions, REST APIs, an MCP server) and stateful orchestration, with human validation before any irreversible action
• Evaluate the agent with a test set, execution traces and written criteria, then fix regressions
• Deploy the agent in a container behind an API, connect it to a channel (WhatsApp Business API or web) and log every action

Audience

Developers, data engineers, software engineers, architects; technical teams who will take over an agent in production.

Prerequisites

Programming in Python (functions, classes, virtual environments, packages), consuming a JSON REST API, using Git. Verified by an online placement test before enrolment. Equipment: one laptop per participant, 16 GB of memory recommended, with Python 3, Docker, Git and a code editor installed before the first day (procedure provided with the joining instructions); a training cloud environment, model API keys and a test WhatsApp number provided by Yolea, credits included; a training Git repository; for remote attendance, a stable connection of at least 10 Mbit/s, microphone and camera; a training corpus is provided, and the participant's own corpus is accepted if anonymised.

Programme

• Integrating model APIs
• Designing document search systems
• Agent orchestration and tool use
• Evaluation and observability
• Containerised deployment and connection to a messaging channel
• Tools for this edition: LangGraph or Pydantic AI for orchestration, LlamaIndex or LangChain for document search, MCP for tools, LangSmith for traces, FastAPI and Docker for deployment, Claude, GPT, Gemini or an open model via Ollama depending on the client's constraints

Format

5 days, 35 hours. In-company or inter-company; on site recommended, hybrid possible; 8 to 12 participants. 5 consecutive days of 7 hours, or 3 + 2 days one week apart with coaching between the modules. 20% input, 80% guided building, each participant developing their own agent end to end; daily code reviews.

Assessment

Beforehand, an online placement test. During the course, a daily check quiz and code review. At the end, a demonstration of the agent to a Yolea jury, assessed on a grid (operation, search quality, safety of actions, logging, deployment), followed by an end-of-course evaluation and a personal certificate of attendance. At three weeks, a follow-up check-in on the move to production.