Lab Tests

This lab immerses you in the practical reality of working with Generative AI tools at a production level. You'll move from simple prompt-and-response to systematic experimentation — testing how different phrasings, system messages, and model parameters (temperature, top_p) produce meaningfully different outputs.

Using both ChatGPT (OpenAI) and Groq's fast inference API, you'll compare output quality, speed, and cost across platforms for the same task — exactly the kind of decision engineering teams face when selecting AI infrastructure.

This lab takes you from talking to AI to building AI that acts. Using n8n (a visual workflow automation tool) and Llama (Meta's open-source LLM family), you'll design and deploy an AI agent that can execute a multi-step workflow autonomously — without you pressing each button.

The shift from "GenAI as a chat tool" to "AI as an automated worker" is one of the most significant in the field. This lab gives you hands-on experience with that shift, including the debugging, edge cases, and design decisions that come with building agentic systems.

This is where everything from Sessions 1–8 converges. You'll build a complete, end-to-end AI application — not a demo, not a single prompt, but a system with distinct components that work together: a retrieval layer, a generation layer, structured output parsing, and basic error handling.

The hiring AI running project provides a natural template: a system that takes a candidate profile and a job description, retrieves the most relevant criteria via RAG, generates a structured assessment, and returns a formatted, machine-readable result. You may adapt this to any domain you choose.

The final lab is a capstone engineering design challenge. You'll design a complete production AI system — from architecture and data pipeline through to deployment strategy, monitoring plan, safety audit, and cost model. This is what senior AI engineers and architects do before a single line of code is written.

Your design will cover: model selection rationale, infrastructure choices, scaling strategy (to 100K+ requests/day), caching plan, safety and bias testing approach, and cost optimisation. You'll also identify failure modes and document your mitigation strategies.

In this lab you will build a fully connected personal AI assistant from scratch. The system uses Telegram as the user-facing interface, n8n as the automation and routing backbone, Llama 3.2 3B running locally via Ollama for language understanding, ChromaDB as a persistent vector memory store, and Gmail for taking real-world actions.

Unlike previous labs that focus on a single component, this one is about integration — making five tools work together reliably. You'll handle intent routing (question vs. remember vs. search vs. email), debug inter-service communication, and experience the design tradeoffs of local-first AI versus cloud APIs. An interactive prototype is included to let you visualise the live pipeline before you build it.

In this lab you will create a Word Counter Agent as an Apify Actor. The actor accepts a web page URL and a word to search for, crawls the page, counts the word, and returns useful context about what it found.

You will work in Apify's browser-based Web IDE, update the actor's JavaScript source and configuration files, and deploy it to the cloud. The result is a reusable web agent that produces a count, page title, case-sensitive comparison, match positions, and sample context from any page.