How carlosom21/self-rag works
EvidenciaLocal is a local, PDF-grounded RAG application with Baseline and Corrective query modes. Users index PDFs, then ask questions through the CLI or Streamlit and receive answers, citations, evidence, and abstentions when needed. Corrective mode critiques evidence and may reformulate retrieval before a single final answer; evaluation supports comparing modes and human annotation. The graph follows documented workflows and excerpted calls; several evaluation and interface details are documented but their implementations were not sampled.
Main entry points: rag_pipeline.py RAG orchestrator · indexer.py Indexing orchestrator · main.py CLI commands · streamlit_app.py Streamlit app
Components 14
User interfaces
- CLI commands
src/evidencialocal/cli/main.py - Streamlit app
src/evidencialocal/ui/streamlit_app.py
Document indexing
- Indexing orchestrator
src/evidencialocal/pipeline/indexer.py - PDF processing
src/evidencialocal/retrieval/document_processor.py
RAG query
- RAG orchestrator
src/evidencialocal/pipeline/rag_pipeline.py - Retriever
src/evidencialocal/retrieval/retriever.py
Local inference and state
- Configuration
src/evidencialocal/config/settings.py - Domain models
src/evidencialocal/core/models.py - E5 embeddings
src/evidencialocal/embeddings/embedder.py - Vector index
src/evidencialocal/vectorstore/chroma_store.py - Metadata and traces
src/evidencialocal/metadata/sqlite_store.py - Local LLM client
src/evidencialocal/llm/llama_client.py
Evaluation
- Mode evaluation
src/evidencialocal/cli/evaluate.py
Other components
- User
Connections 18
- User → CLI commands: runs commands
README.md - User → Streamlit app: uses interface
README.md - CLI commands → Indexing orchestrator: indexes PDFs
src/evidencialocal/cli/main.py - CLI commands → RAG orchestrator: submits questions
src/evidencialocal/cli/main.py - Streamlit app → Indexing orchestrator: indexes uploads
src/evidencialocal/ui/streamlit_app.py - Streamlit app → RAG orchestrator: submits questions
src/evidencialocal/ui/streamlit_app.py - Indexing orchestrator → PDF processing: processes PDFs
src/evidencialocal/pipeline/indexer.py - PDF processing → E5 embeddings: embeds chunks
src/evidencialocal/retrieval/document_processor.py - Indexing orchestrator → Vector index: stores vectors
src/evidencialocal/pipeline/indexer.py - Indexing orchestrator → Metadata and traces: stores metadata
src/evidencialocal/pipeline/indexer.py - RAG orchestrator → Retriever: requests retrieval
src/evidencialocal/pipeline/rag_pipeline.py - Retriever → E5 embeddings: embeds queries
src/evidencialocal/retrieval/retriever.py - Retriever → Vector index: queries vectors
src/evidencialocal/retrieval/retriever.py - RAG orchestrator → Local LLM client: critiques and generates
README.md - RAG orchestrator → Metadata and traces: records traces
src/evidencialocal/pipeline/rag_pipeline.py - CLI commands → Mode evaluation: runs evaluation
src/evidencialocal/cli/main.py - Mode evaluation → RAG orchestrator: compares modes
src/evidencialocal/cli/evaluate.py - Mode evaluation → Metadata and traces: stores results
src/evidencialocal/cli/evaluate.py