How butlee27/advanced-conversational-rag works
This is a Streamlit conversational RAG application: users upload PDFs, which are indexed, then ask questions answered from retrieved document context with session memory and safety checks. Retrieval combines semantic and BM25 search, followed by reranking; the app checks groundedness, sanitizes output, and displays source pages. Evaluation with Ragas is a separate capability. The graph follows README documentation and indexed or excerpted imports/calls; unsampled implementation details and connections are omitted where evidence is limited.
Main entry points: app.py Streamlit Interface · indexer.py Index Builder · history_retriever.py History-Aware Search · chains.py RAG Chain
Components 24
Application
- Streamlit Interface
app.py - Safety and Grounding
src/guardrails.py - Answer and Sources
app.py
Document Indexing
- PDF Loader
src/loader.py - Index Builder
src/indexer.py - Text Splitter
src/splitter.py - Embeddings
src/embeddings.py - Chroma Store
src/vectordb.py
Retrieval and Memory
- Conversation Handler
src/conversation.py - Session History
src/history.py - History-Aware Search
src/history_retriever.py - Advanced Retriever
src/custom_retriever.py - Hybrid Search
src/hybrid_retriever.py - Vector Search
src/retriever.py - BM25 Search
src/bm25.py - Cross-Encoder Reranker
src/reranker.py
Answer Generation
- RAG Chain
src/chains.py - RAG Prompts
src/prompts.py - Groq Language Model
src/llm.py
Evaluation
- Evaluation Runner
eval/run_evaluation.py - Evaluation Dataset
eval/dataset.py - Ragas Metrics
eval/evaluate_ragas.py
Other components
- PDF User
- Evaluation Operator
Connections 32
- PDF User → Streamlit Interface: uploads and asks
README.md - Streamlit Interface → PDF Loader: checks PDFs
app.py - Streamlit Interface → Index Builder: builds index
app.py - Index Builder → PDF Loader: loads documents
src/indexer.py - Index Builder → Text Splitter: splits documents
src/indexer.py - Index Builder → Embeddings: creates vectors
src/indexer.py - Index Builder → Chroma Store: stores chunks
src/indexer.py - Chroma Store → Embeddings: embeds documents
src/vectordb.py - Streamlit Interface → Conversation Handler: creates conversation
app.py - Streamlit Interface → Safety and Grounding: checks and sanitizes
app.py - Streamlit Interface → Groq Language Model: checks grounding
app.py - Streamlit Interface → Answer and Sources: renders answer and sources
app.py - Conversation Handler → RAG Chain: wraps chain
src/conversation.py - Conversation Handler → Session History: loads session history
src/conversation.py - RAG Chain → Groq Language Model: generates answer
src/chains.py - RAG Chain → RAG Prompts: uses prompt
src/chains.py - RAG Chain → History-Aware Search: retrieves context
src/chains.py - History-Aware Search → Groq Language Model: rewrites question
src/history_retriever.py - History-Aware Search → RAG Prompts: uses rewrite prompt
src/history_retriever.py - History-Aware Search → Advanced Retriever: configures retriever
src/history_retriever.py - Advanced Retriever → Hybrid Search: searches candidates
src/custom_retriever.py - Advanced Retriever → Cross-Encoder Reranker: reranks results
src/custom_retriever.py - Hybrid Search → Vector Search: combines search
src/hybrid_retriever.py - Hybrid Search → BM25 Search: combines search
src/hybrid_retriever.py - Vector Search → Chroma Store: searches vectors
src/retriever.py - BM25 Search → PDF Loader: loads documents
src/bm25.py - BM25 Search → Text Splitter: uses chunks
src/bm25.py - Evaluation Operator → Evaluation Runner: runs evaluation
README.md - Evaluation Runner → Evaluation Dataset: loads dataset
eval/run_evaluation.py - Evaluation Operator → Ragas Metrics: evaluates RAG metrics
README.md - Ragas Metrics → Groq Language Model: uses model
eval/evaluate_ragas.py - Ragas Metrics → Embeddings: uses embeddings
eval/evaluate_ragas.py