ChandraPrakashBathula/Self-RAG

Naive RAG vs Agentic RAG

An interactive playground for self-correcting retrieval augmented generation. Build a naive RAG pipeline over your own document, then run the same question through a reflect, verify, retry loop and watch every decision.

Configuring this Space

Set these under Settings, Variables and secrets:

Without a key the Space starts but every generation call fails, because a Space cannot run a local Ollama server. The fully local path documented in the main README works when you clone the repository and run it yourself.

What differs from a local clone

• —Cloud models only. Ollama needs a model server on the same machine, which a Space does not provide. Clone the repo for the offline path.
• —Session state is in memory and per browser tab. A Space that sleeps or restarts loses uploaded documents and indexes. Rerun the wizard from Step 1.
• —One process. Do not add --workers; pipeline state is not shared across workers.
• —Your key funds every visitor. A public Space spends your Groq quota on strangers. Consider duplicating it privately, or asking visitors to supply their own key.
• —The full textbook is not bundled here. The worked example runs from a pre-extracted Chapter 3-6 excerpt that ships with the code. The complete 1,208-page Human_Nutrition.pdf lives in the GitHub repository instead, to keep this image small; download it there to follow Levels B and C of the exercises.

Rename before use

Hugging Face expects the Space README at README.md. This repository already has one for the project itself, so rename this file when you create the Space.