How shruti-06/cuisine-knowledge-rag-system works
This repository is a notebook-driven South Asian cuisine RAG application. A user submits a free-text question or JSON batch through Gradio; questions are retrieved against the prepared corpus, reranked, and answered by a grounded language model, with retrieved context returned to the user. The notebook also documents corpus preparation, chunking, embedding, and benchmark evaluation. Relationships follow the README’s documented pipeline; source code was not excerpted, so implementation-level call ownership cannot be verified.
Main entry points: Codebase.ipynb Dense retrieval · bge_chunks.pkl Chunk store · Background_Corpus.json Cleaned corpus · bge_index.faiss FAISS vector index
Components 21
Corpus preparation
- Wikipedia sources
- Wikibooks sources
- Food blog sources
- Wikipedia collection
Notebook/Codebase.ipynb - Wikibooks collection
Notebook/Codebase.ipynb - Blog collection
Notebook/Codebase.ipynb - Cleaned corpus
Data/Background_Corpus.json - Adaptive chunking
Notebook/Codebase.ipynb
Retrieval
- Chunk store
Indexes/bge_chunks.pkl - BGE embeddings
Notebook/Codebase.ipynb - FAISS vector index
Indexes/bge_index.faiss - Dense retrieval
Notebook/Codebase.ipynb - BM25 retrieval
Notebook/Codebase.ipynb - Hybrid fusion
Notebook/Codebase.ipynb - Cross-encoder reranking
Notebook/Codebase.ipynb
Answering and evaluation
- Grounded generation
Notebook/Codebase.ipynb - Benchmark dataset
Data/RAG_Benchmark_Dataset.json - Retrieval and answer evaluation
Notebook/Codebase.ipynb
User interface
- Gradio interface
Notebook/Codebase.ipynb - JSON batch route
Notebook/Codebase.ipynb
Other components
- User
Connections 26
- Wikipedia sources → Wikipedia collection: provides pages
README.md - Wikibooks sources → Wikibooks collection: provides pages
README.md - Food blog sources → Blog collection: provides posts
README.md - Wikipedia collection → Cleaned corpus: cleans and collects
README.md - Wikibooks collection → Cleaned corpus: cleans and collects
README.md - Blog collection → Cleaned corpus: cleans and collects
README.md - Cleaned corpus → Adaptive chunking: chunks documents
README.md - Adaptive chunking → Chunk store: stores chunks
README.md - Chunk store → BGE embeddings: embeds chunks
README.md - BGE embeddings → FAISS vector index: builds index
README.md - FAISS vector index → Dense retrieval: supplies vectors
README.md - Chunk store → Dense retrieval: provides chunks
README.md - Chunk store → BM25 retrieval: provides chunks
README.md - Dense retrieval → Hybrid fusion: contributes scores
README.md - BM25 retrieval → Hybrid fusion: contributes scores
README.md - Hybrid fusion → Cross-encoder reranking: ranks candidates
README.md - Cross-encoder reranking → Grounded generation: provides context
README.md - Benchmark dataset → Retrieval and answer evaluation: supplies queries
README.md - Dense retrieval → Retrieval and answer evaluation: evaluates retrieval
README.md - Grounded generation → Retrieval and answer evaluation: evaluates answers
README.md - User → Gradio interface: submits questions
README.md - User → JSON batch route: uploads queries
README.md - JSON batch route → Gradio interface: routes batch input
README.md - Gradio interface → Dense retrieval: retrieves context
README.md - Gradio interface → Grounded generation: requests answers
README.md - Grounded generation → Gradio interface: returns grounded answer
README.md