RAG MicroSim
This MicroSim lets you experiment with Retrieval-Augmented Generation end to end. Type a question, choose a document corpus, and run the three RAG steps. You will see which documents are retrieved (with relevance scores), how the augmented prompt is assembled from color-coded sections, and what response the system generates.
Interactive Demo
To embed this MicroSim in your own page, use the following iframe:
1 | |
Overview
The simulator is divided into three working areas:
- Top: a query box and a corpus selector (Company Policies, Product Docs, or HR Handbook). Selecting a corpus loads a sample question.
- Left panel: the top-K retrieved documents, each with a green relevance bar and a numeric score. The best match is highlighted.
- Right panel: the augmented prompt built from three color-coded sections -
[System]in purple,[Context]in orange,[Query]in blue - followed by the generated response.
A three-step progress indicator shows which stage is running. Pressing Run RAG Process animates through Retrieval, Augmentation, and Generation in sequence.
Controls:
- K slider (1-10): how many documents to retrieve.
- Temperature slider (0-1): simulated generation variability; it changes how the response is phrased.
- Include sources in prompt checkbox: toggles whether document titles are cited.
- Run RAG Process and Reset buttons.
Retrieval uses keyword-overlap scoring as a stand-in for cosine similarity over embeddings, and generation is template-based - no real model is called - so the behavior is deterministic and easy to reason about.
Lesson Plan
- Vary K: Set K to 1, then to 5, and discuss how more documents change the context (and the risk of irrelevant material).
- Inspect the prompt: Have students point to the
[System],[Context], and[Query]sections and explain each one's job. - Toggle sources: Turn off "Include sources" and discuss what is lost for citations and trust.
- Different queries: Ask the same question against different corpora and observe how retrieval scores shift.
- Temperature: Compare a low-temperature and high-temperature response and relate it to factual reliability.