Completed workshop

From Prompt to Agent: Build Your First Intelligent AI System

The workshop introduced practical agentic AI through Python, LLM calls, Streamlit interfaces, system prompts, MCP tools, RAG, robotics agents, recipe agents, and a final RobotChef challenge.

18 April 2026

10:00 - 18:00

College Lane, University of Hertfordshire

University of Hertfordshire

GitHub repository

Exercises and code

Live recording

Watch on YouTube

Workshop story

From a first LLM call to a working multi-agent system.

A full-day hands-on workshop where students moved from their first LLM API call to a working multi-agent RobotChef platform.

Open GitHub Watch live recording Request slides

Connected Python code to an LLM backend

Built a Streamlit chatbot interface

Used system prompts to shape expert behaviour

Created MCP-powered robotics and recipe agents

Combined agents into a multi-agent RobotChef platform

Completed a budget and nutrition challenge

Why it mattered

From AI theory to evidence students can show.

The workshop connected MSc-level computing knowledge with the practical patterns now expected in applied AI engineering roles.

The role has shifted

Students saw why employers increasingly look for AI engineers who can turn models into usable software, not just explain AI concepts.

Portfolio evidence matters

The day was designed around visible proof: code, a working interface, and an agent workflow students can discuss and extend.

Agentic AI is already practical

Tools, memory, loops, RAG, and multi-agent coordination were presented as engineering patterns students can reuse after the event.

Skills covered

PythonAPIsPrompt EngineeringLLMsMCP ToolsAgent LoopsStreamlitGitHubRAGMulti-Agent Systems

Learning journey

A practical path from prompt to working system.

Each stage added one new capability: interface, behaviour, tools, retrieval, then multi-agent coordination.

LLM API call

Chatbot UI

System prompt

MCP robotics agent

Recipe agent

Multi-agent RobotChef

Build sessions

Agents that reason, use tools, and collaborate.

Students worked through increasingly capable systems: a tool-using robotics designer, a recipe analysis agent, and an integrated RobotChef platform.

Robotics Agent

An MCP-powered agent that explores component, sensor, and actuator data, tests tools directly, then runs a full robotics platform designer.

Recipe Agent

A culinary analysis agent that browses dish data, tests recipe tools, and reasons about ingredients, techniques, equipment, safety, and feasibility.

RobotChef A2A Platform

A Streamlit platform where the recipe agent and robotics agent work together as an agent-to-agent workflow.

Repository-based curriculum

What students actually built from the GitHub repo.

The workshop code was organised into self-contained sessions, so participants could follow the build step by step and revisit each module later.

Session 1

LLM foundations and chatbot UI

First LLM call

Multi-turn conversation

Temperature comparison

Streamlit chatbot

Persona prompts

Session 2

MCP and robotics tools

Explore robotics data

Test MCP tools directly

Run the robotics agent

Session 3

RAG concepts

Retrieval demo

Streamlit RAG interface

Grounding answers with retrieved context

Session 4

Recipe agent

Explore dish data

Test recipe tools

Run culinary analysis agent

Session 5

Agent-to-agent platform

Connect agents

Run the RobotChef app

Display an end-to-end AI workflow

Final team exercise

Smart Budget RobotChef Challenge was used as a short closing exercise: The day ended with a short team extension task where participants adapted RobotChef around budget, nutrition, and usability.

Agenda