How hocine-drioueche/randoneo-rag works
Randoneo RAG is a French-language customer-support assistant grounded in an outdoor-shop document corpus. The main workflow loads Markdown, splits it into overlapping chunks, indexes embeddings in Chroma, retrieves four passages per question, and generates a context-bound answer with source citations and a refusal when evidence is missing. chat.py exposes this workflow interactively; evaluate.py replays it against question sets and scores retrieval and generation with RAGAS. The README documents the stages; unsampled corpus and evaluation data are represented by their exact directories.
Main entry points: chat.py Interactive chat · evaluate.py Evaluation runner · rag.py RAG chain · corpus Shop documents
Components 15
Chat interaction
- Interactive chat
chat.py - Cited answer
chat.py
RAG pipeline
- Corpus loader
rag.py - Chunk splitter
rag.py - OpenAI embeddings
- Chroma index
rag.py - Passage retriever
rag.py - RAG chain
rag.py - Grounded prompt
rag.py - Claude model
Knowledge corpus
- Shop documents
corpus
Evaluation
- Evaluation runner
evaluate.py - Question sets
eval - RAGAS metrics
Other components
- Questioner
Connections 19
- Questioner → Interactive chat: asks
README.md - Interactive chat → Corpus loader: loads corpus
chat.py - Corpus loader → Shop documents: reads Markdown
rag.py - Interactive chat → Chunk splitter: splits documents
chat.py - Interactive chat → Chroma index: builds index
chat.py - Chroma index → OpenAI embeddings: vectorizes chunks
README.md - Interactive chat → Passage retriever: configures retrieval
chat.py - Passage retriever → Chroma index: searches passages
README.md - Interactive chat → RAG chain: assembles chain
chat.py - RAG chain → Passage retriever: retrieves context
rag.py - RAG chain → Grounded prompt: formats context
rag.py - RAG chain → Claude model: generates answer
rag.py - Interactive chat → Cited answer: displays response
chat.py - Cited answer → Questioner: returns answer
README.md - Evaluation runner → Corpus loader: builds RAG
evaluate.py - Evaluation runner → Question sets: loads questions
evaluate.py - Evaluation runner → Passage retriever: retrieves contexts
evaluate.py - Evaluation runner → RAG chain: replays questions
evaluate.py - Evaluation runner → RAGAS metrics: scores samples
evaluate.py