WordSwarm AI: Reasoning & Benchmarking Workshop

Welcome to the WordSwarm AI workshop!

The live demo is a WordSwarm game with an AI agent that can play it autonomously.

WordSwarm AI Demo

What you’ll learn

In this workshop, you will:

  • Play with reasoning models — use /think and /no-think prompting to control LLM reasoning behavior in a live AI agent

  • Benchmark MaaS models — use GuideLLM to measure throughput, latency, and concurrency of models served via vLLM on OpenShift

  • Compare model performance — see how different models (kimi-k2-5, Nemotron, Llama, Gemma) perform on real-time tasks vs. synthetic benchmarks

The game

WordSwarm is a honeycomb word game (originally Intel Corp, 2012) with an AI agent bolted on. The agent plays autonomously against a ticking clock — it enumerates adjacency paths through a 17-cell hex grid, matches them against a dictionary, and uses an LLM to handle ambiguous cases.

WordSwarm AI Dashboard

Who this is for

This workshop is designed for AI/ML engineers, platform engineers, and developers who want hands-on experience with:

  • LLM reasoning capabilities (/think, /no-think)

  • Model benchmarking with GuideLLM

  • Model-as-a-Service (MaaS) on OpenShift with vLLM

Experience level: Beginner to Intermediate

Prerequisites

  • A terminal with curl and pip installed

  • Access to the MaaS API endpoint

  • A valid $TOKEN for the MaaS API

Estimated time

This workshop takes approximately 30 minutes to complete.

Let’s get started!

Click on Workshop Overview in the navigation to begin.