Mini RAG search
Search a small in-browser corpus and inspect retrieved chunks
Interactive demo · mock data. The “score” is a simple word-overlap fraction, not an embedding. The documents are invented handbook snippets. Nothing is sent to a model or a database.
Account logs are kept for 90 days. Uploaded documents are deleted 14 days after a workspace is closed. This demo does not describe a real legal requirement.
Keep a small evaluation set of questions with expected answers. Re-run it when the prompt, the model or the retrieved documents change, and record pass or fail by hand at first.
Refunds are available within 30 days of purchase if the software licence key has not been activated. Requests go to billing@example.com. Chargebacks are reviewed case by case.
Account logs are kept for 90 days. Uploaded documents are deleted 14 days after a workspace is closed. This demo does not describe a real legal requirement. Keep a small evaluation set of questions with expected answers. Re-run it when the prompt, the model or the retrieved documents change, and record pass or fail by hand at first.
About this demo
Type a question and retrieve the closest chunks from a fixed set of demo documents, using a simple overlap score (not a real embedding model). It previews the retrieve-then-read flow I write about when comparing pgvector and hosted vector stores. The answers are stitched from those chunks only.
Features
- A fixed demo corpus (no upload, no server)
- Overlap score, honestly labeled as not a real embedding
- Retrieved chunks shown before the drafted answer
- Runs entirely in your browser
Tech used
Use cases
- Explaining retrieval before generation
- A starting point for an evaluation set
Want the production version?
Real contracts, APIs and integrations, scoped to your project.