ALEX Institute / AI / AI Labs

AI Labs

Hands-on AI for beginners. Each lab course is 6 hours: 2 hours of concept and 4 hours of guided lab. Everything runs on an AI computer in the classroom, so there are no API keys, no cloud bills and no subscriptions. Student cost: $0.

Why AI Labs

Built for beginners

For students with little technical background. Visual logic, clear concepts and hands-on builds instead of dense computer-science jargon.

6-hour lab courses

Each course is 2 hours of concept and 4 hours of guided lab. You leave every course with something you built.

$0 and open

No charge for software, API keys, cloud compute or datasets. Open-source tools, open-source models and openly licensed data.

An AI computer in the room

Students use laptops and classroom computers connected to an NVIDIA DGX Spark system (GB10 Grace Blackwell Superchip, 128 GB unified memory). Models, agents and fine-tuning run on the classroom network.

Learning path

Three AI phases

AI Labs is the hands-on, in-classroom way to learn. The self-paced AI Classes 1–4 and tracks stay open to everyone online. Quantum AI lab work belongs to the Quantum Computing section, starting with Quantum Leadership.

Phase 1 · 12 hours

Core AI Foundations

Learn the habits and the clean data that let you guide open AI models accurately, without made-up answers.

Lab 1A

Structured Prompting with Open Models

6 hours total (2h concept | 4h guided lab) · Student cost: $0

Modules

  1. How AI models predict text: tokens, context windows, and word probability versus human reasoning
  2. Prompting without made-up answers: role-based instructions, strict formatting rules and few-shot examples
  3. Comparing open models: the same prompts across IBM Granite, Mistral Small and Qwen2.5, all running on the classroom NVIDIA DGX Spark system

Open data used in the labs

US Government public-domain data: NOAA weather records and federal service logs from Data.gov. Each dataset's license is checked before a lab uses it.

Open toolkit

  • AnythingLLM, hosted on the DGX Spark system (MIT)
  • Ollama (MIT)
  • IBM Granite 3.3, Mistral Small 3.1 and Qwen2.5 (all Apache-2.0)

Capstone build

A prompt template that turns raw government logs into structured summaries, with no made-up content.

Lab 1B

Data Hygiene and Table Formatting

6 hours total (2h concept | 4h guided lab) · Student cost: $0

Modules

  1. Why messy data breaks AI: duplicates, broken formatting and missing fields
  2. Spreadsheet cleanup: turning raw text rows into clean tables with open-source tools
  3. Machine-readable formats: clean CSV and JSON that an AI can read automatically

Open data used in the labs

The CFPB Consumer Complaint Database: real, messy consumer complaint records published by the US Government (public domain).

Open toolkit

  • LibreOffice Calc (MPL-2.0)
  • OpenRefine (BSD-3)
  • AnythingLLM on the DGX Spark system

Capstone build

A cleaned, validated CSV or JSON dataset, paired with a prompt that parses it automatically.

Phase 2 · 18 hours

Agents and Agent Teams

Go from one-off chat prompts to building digital workers ("agents") that work from verified documents, then link them into teams that check their own work. See also Agentic AI.

Lab 2A

Single AI Workers on a Local AI Computer

6 hours total (2h concept | 4h guided lab) · Student cost: $0

Modules

  1. From chatbots to agents: the ReAct loop for tasks with several steps
  2. No API keys needed: pointing open-source agent frameworks at the DGX Spark system's local Ollama endpoint instead of paid API keys or cloud accounts
  3. Connecting tools: safe access to local folders, calculators and offline search tools

Open data used in the labs

Research papers from arXiv released under a CC BY license only, each credited by title, author, source and license.

Open toolkit

  • CrewAI (MIT)
  • Ollama local API on the DGX Spark system
  • Qwen2.5-Coder 7B, 14B or 32B (Apache-2.0)
  • IBM Granite 3.3 (Apache-2.0)

Capstone build

A research worker that reads local arXiv papers on its own and writes structured executive briefs.

Lab 2B

Grounded Memory and Citations (RAG)

6 hours total (2h concept | 4h guided lab) · Student cost: $0

Modules

  1. Stopping guesswork with RAG: answering strictly from verified documents
  2. Chunking and embeddings: turning long PDFs into searchable vectors with IBM Granite open embedding models
  3. Checking citations: an exact page and paragraph citation on every answer

Open data used in the labs

Public-domain technical handbooks and historical reference texts from Project Gutenberg, plus Wikibooks texts (CC BY-SA, credited by title, author, source and license).

Open toolkit

  • ChromaDB (Apache-2.0)
  • AnythingLLM (MIT)
  • IBM Granite embedding models (Apache-2.0) on the DGX Spark system

Capstone build

A private question-and-answer assistant that cites its sources, grounded in public-domain technical manuals.

Lab 3

Multi-Agent Teams and Self-Correction

6 hours total (2h concept | 4h guided lab) · Student cost: $0

Modules

  1. Designing a digital department: Researcher → Writer → Fact-Checker
  2. Handoffs between agents: Microsoft AutoGen (MIT) and LangGraph (MIT) passing structured checklists and drafts
  3. Self-correction loops: automatic critic rules that catch formatting or factual errors and force a retry without a person stepping in

Open data used in the labs

An offline copy of Wikipedia (Kiwix, CC BY-SA, credited by title, author, source and license) stored on the DGX Spark system, so the Fact-Checker agent checks claims fully offline.

Open toolkit

  • Being written.

Capstone build

Being written.

Phase 3 · 18 hours

Infrastructure and Sovereign AI

Three lab courses on running AI you control. The lab courses are being written. Until then, the openly licensed readings in Track D · AI Sovereignty cover who controls AI: data, models, rules and the machines it runs on.

Open by design

Every tool and model in AI Labs is under an OSI-approved open-source license (MIT, Apache-2.0, BSD-3 or MPL-2.0). Data is US Government public domain or openly licensed, and we check each dataset's license before a lab uses it. Openly licensed text is credited by title, author, source and license. ALEX Institute is not affiliated with or endorsed by NVIDIA, IBM, Mistral AI, Alibaba Cloud (Qwen), Microsoft or any other provider named here.


ALEX Institute