Miatz playground

RAG Sandbox

See why that chunk got retrieved, in 2D space.

Coming to the Build-LabAIE-202BYOK AIAI-Coredev · data

The Demystify signature

The 2D embedding plot makes semantic similarity a literal picture instead of a metaphor — learners watch their own chunk get pulled toward or away from the query as they edit the question.

How it works

What goes in, what comes out

What it does

Learner uploads a document set, picks a chunking strategy, and asks a question; instead of just returning an answer, the sandbox projects every chunk's embedding into a 2D plot (pgvector-backed similarity) with the query as a labeled point, so retrieval becomes a literal spatial picture. A citation-truth overlay then highlights, span by span, exactly which retrieved chunk each sentence of the model's answer maps back to — or flags it unsupported.

You bring

A document set (PDF/markdown/text), a natural-language question, and an optional known-answer for scoring.

You get

A 2D embedding plot of chunks vs. query, the retrieved top-k with scores, the generated answer with a citation-truth overlay, and a retrieval-quality score.

You control

Chunk size/overlap, embedding model, top-k, hybrid (keyword+vector) weighting, reranker on/off, and citation-strictness (refuse below a confidence threshold).

Where it's used

In-module lab for AIE-202/203, graded eval-gate lab (retrieval score at or above threshold); reused as a grounding exercise inside RAG-heavy specializations.

chunking tradeoffsembeddings & vector similarityhybrid search + rerankinglost-in-the-middlecitation grounding

Optional · watch & try

Go deeper, elsewhere

Hand-picked public explainers and open tools that complement this one — always optional, never required, never graded.

Retrieval Augmented Generation (RAG) Explained in 8 Minutes!Fast end-to-end mental model of the RAG pipeline · ~8 min RAG Explained in 12 MinutesSecond angle on the same pipeline with a different worked example · ~12 min RAG Explained: What Is Retrieval Augmented Generation?Explains why RAG beats relying on the model's parametric memory alone RAG Playground - Interactive RAG Pipeline VisualizationStep through an actual retrieval-then-generation pipeline interactively Vector Databases Explained in 100 SecondsFastest possible primer on what a vector database is for · ~2 min Vector Database Explained in under 3 MinutesA second short framing for reinforcement · ~3 min

Concepts click when you open the machinery.

Three labs are already live and free — the same hands-on style this playground brings to its module.

Play the free labs Take the DSAT