Building Intelligent Agents

The course is aimed at professionals with basic programming experience who want to understand and build practical AI agents using large language models (LLMs). It is suitable for professionals working in data science, and related disciplines, as well as practitioners seeking to apply agent-based approaches to real-world problems. Familiarity with Python is expected; prior experience with LLMs or agent frameworks is helpful but not required.

Course Requirements

Prerequisites:

• Python programming (intermediate level)
• Command line comfort
• Git basics

Tools & Platforms:

• Python 3.10+
• LangChain, LlamaIndex
• Ollama or similar local LLM runtime
• GitHub for project management

After the course participants will be able to:

• Explain the agent loop and distinguish an agent from a fixed LLM workflow.
• Implement tool calling and API integration with validation and error handling.
• Build a small retrieval-grounded application.
• Evaluate behaviour systematically and diagnose common failure modes.
• Apply basic security, privacy and human-oversight principles.
• Justify design choices in a working agent prototype.

Module 1. Agent fundamentals and LLMs
Topics: Agent loop and core concepts; LLM capabilities and limitations; prompts and structured outputs; choosing a model/API. 
Practical work: Build a minimal Python agent for a clearly defined task; compare prompts and define simple success tests. 

Module 2. Tool use and agent control flow 
Topics: Function/tool calling; schema validation; API integration; sequential and conditional workflows; retries, errors and stopping conditions. 
Practical work: Extend the agent with two tools and robust input/output validation. Introduce frameworks only where they simplify implementation. 

Module 3. Retrieval and knowledge grounding 
Topics: RAG fundamentals: document preparation, chunking, embeddings and retrieval; source-grounded responses; conversation state versus persistent memory. 
Practical work: Build a document Q&A; assistant that retrieves evidence and identifies its sources. Compare retrieval with prompting alone. 

Module 4. Agent patterns, evaluation and safety 
Topics: Workflow versus single-agent versus multi-agent designs; lightweight planning and ReAct-style tool use; task-specific evaluation; tracing and failure analysis; prompt injection, permissions and human approval. 
Practical work: Evaluate the agent on a small test set, inspect failures, and add appropriate safeguards. Multi-agent orchestration is discussed selectively, not built as a separate system. 

Module 5. Deployment essentials and capstone project 
Topics: Practical deployment choices; latency and token-cost trade-offs; secrets and access control; monitoring and regression tests; communicating limitations. 
Practical work: Demonstrate the capstone, report evaluation results, and explain architecture, failure modes and trade-offs. 

Example Projects:

• Personal research assistant with web access
• Code review agent for specific frameworks
• Customer support agent for a domain
• Data analysis agent with visualization tools
• Educational tutor for technical subjects

You can find an overview of our continuing education programmes in the field of computational science and artificial intelligence here.

The course is strongly based on applied exercises. In addition to traditional lectures, the participants will experience:

• parctical programming exercises
• group work
• self-study (preparation and follow-up)
• final capstone project

Participants will have to submit and present a final capstone implementation project. There are 5 lessons organized once a week on Wednesday's afternoon. After 4 lessons the participants work on their final project for about two weeks. The final project presentation will be held during the last lesson

Successful completion of the course and the assessment will result in a ZHAW microcredential certificate being issued. This digital, verifiable form of certification makes your acquired competencies visible.

Participants who attend the course but do not complete the assessment will receive a course certificate without ECTS credits.