mohamed-amine-sassi/selfcorrectingrag

How mohamed-amine-sassi/selfcorrectingrag works

This repository implements PDF-based retrieval-augmented question answering. Its main workflow extracts and chunks a PDF, embeds the chunks in a persistent Chroma collection, then invokes a correction graph that retrieves and grades context, generates an answer, and may rewrite the query or regenerate an answer. The graph’s shared state is a useful internal layer. The tree also includes a RAGAS evaluation area; its evaluator’s implementation was read but not excerpted, so its connections are not shown. Edges below are limited to relationships visible in excerpts or the source index.

Main entry points: Graph.py Correction graph · main.py RAG entry point · Chunking.py Text chunking · Embedding.py Embedding store setup

Components 18

Document ingestion

PDF input files/AIBook2.pdf
RAG entry point RAGBase/main.py
PDF extraction RAGBase/Extracting.py
Text chunking RAGBase/Chunking.py
Embedding store setup RAGBase/Embedding.py

Retrieval and generation

Chroma collection
Similarity retrieval RAGBase/Retrieve.py
Answer generation RAGBase/Generate.py

Answer correction

Correction graph Correction/Graph.py
RAG state Correction/RAGState.py
Retrieve node Correction/RetreiveNode.py
Grade documents Correction/GradeDocuments.py
Routing decisions Correction/DecisionNodes.py
Rewrite query Correction/RewriteQuery.py
Generate node Correction/GenerateNode.py
Grade answer Correction/GradeAnswer.py

Evaluation

RAGAS evaluation RAGAS/Evaluate.py

Other components

Questioner

Connections 17

Questioner → RAG entry point: asks question RAGBase/main.py
PDF input → RAG entry point: input PDF RAGBase/main.py
RAG entry point → Text chunking: chunks PDF RAGBase/main.py
Text chunking → PDF extraction: extracts text RAGBase/Chunking.py
RAG entry point → Embedding store setup: embeds chunks RAGBase/main.py
Embedding store setup → Chroma collection: creates or loads RAGBase/Embedding.py
RAG entry point → Correction graph: invokes graph RAGBase/main.py
Correction graph → Retrieve node: registers node Correction/Graph.py
Correction graph → Grade documents: registers node Correction/Graph.py
Correction graph → Grade answer: registers node Correction/Graph.py
Correction graph → Rewrite query: registers node Correction/Graph.py
Correction graph → Generate node: registers node Correction/Graph.py
Correction graph → Routing decisions: routes via Correction/Graph.py
Retrieve node → Similarity retrieval: calls retrieval Correction/RetreiveNode.py
Similarity retrieval → Chroma collection: searches chunks RAGBase/Retrieve.py
Generate node → Answer generation: calls generator Correction/GenerateNode.py
Correction graph → RAG state: uses state schema Correction/Graph.py

Drawn by GitDiagram on November 12, 2026 from the repository's file tree, README and source files. Read as Markdown ·