AI Labs — see inside the machinery

Five live playgrounds for tokenization, chunking, embeddings, retrieval and sampling.

Don't just read about RAG — run one, over the platform's own content and your own journey
Make abstract concepts tangible: watch text become tokens, meaning become geometry
Try the free, no-signup versions of several labs before you even log in

Run the machinery, don't just read about it

The AI Labs are five live playgrounds for the concepts underneath every modern AI system — running over the platform's own knowledge and your own learning journey, not toy data.

Chunk Lab — watch a document get sliced the way RAG systems do it: character, recursive, sentence and markdown-aware splitters, with chunk size and overlap you control. Try overlap 0 and find the sentence that gets cut in half.
Tokenizer Lab — paste text and see the tokens a model actually reads, why code and non-English scripts cost more, and what a prompt costs. Compare the same sentence in English and Hindi.
Embedding Lab — turn words into vectors and see meaning as geometry: projections, nearest neighbours, similarity scores.
RAG Lab — run keyword vs semantic vs hybrid retrieval side by side, see exactly how the final prompt is assembled, and compare grounded answers with citations against hallucination without context. Ask "what did I struggle with this month?" against your own journals.
Sampling Lab — one prompt at three temperatures. See why extraction pipelines run at 0 and brainstorms don't.

Availability

The Labs are a module your admin enables from the admin portal; until then you'll see a not-enabled panel. Three of the labs (Embedding, RAG, Sampling) also call live model APIs, so they additionally need the platform's AI configured — they show a needs AI config pill otherwise. The Chunk and Tokenizer labs run entirely in your browser.

Free versions, no account needed

Public, no-signup versions of several labs live on the marketing site at /play: the Tokenizer Lab, the Chunking Lab, and Prompt Anatomy — a COSTAR-framework dissection of a production-grade prompt. Share them freely.

The journey, step by step

Step	What you do	What you get
Open the hub	Go to AI Labs and pick one of the five playgrounds	Each card tells you what you'll learn, plus a suggested experiment
Run the experiment	Try the "Try this" prompt on each card	A concrete aha moment — like finding the sentence a bad chunker cuts in half
Go deeper	Change the parameters — chunk size, temperature, retrieval mode	You see cause and effect directly, not from a diagram
Connect it to your work	In the RAG Lab, query your own journals and the platform corpus	You understand how the platform's own AI features are built

Frequently asked

What are the five labs?

Chunk Lab (how documents get sliced for RAG), Tokenizer Lab (how text becomes the tokens a model reads, and what they cost), Embedding Lab (words as vectors, meaning as geometry), RAG Lab (a real retrieval pipeline with keyword vs semantic vs hybrid search), and Sampling Lab (what temperature actually does).

Why do I see "AI Labs isn't enabled yet"?

The Labs are a module your admin switches on from the admin portal. Until then, learners see a friendly not-enabled panel — ask your admin, or try the free /play versions.

Why is a lab marked "needs AI config"?

The Embedding, RAG and Sampling labs call live model APIs, so they need the platform's AI configured by an admin. The Chunk and Tokenizer labs run entirely in your browser.

Is there a version I can share with friends who aren't on Miatz?

Yes — free, no-signup playgrounds live at miatz's public /play page: the Tokenizer Lab, the Chunking Lab and Prompt Anatomy (dissect a production prompt with the COSTAR framework).