How digibhatia/rag-advance works
Disha is a document-grounded Q&A pipeline. Its main workflow loads varied source documents, redacts PII, chunks and embeds the text into a persistent vector store, then serves questions through retrieval, confidence checks, evidence packing, and cited LLM answers. The README also documents an HTTP question client and optional vector-store inspection. Edges reflect calls shown in excerpts or relationships stated in the README; the inspection tool’s internal connections are documented but not independently traced here.
Main entry points: app.py Q&A API · build_vectorstore.py Index Builder · retrieval.py Retrieval Patterns · loaders.py Format Loaders
Components 15
Knowledge Ingestion
- Source Documents
documents - Index Builder
build_vectorstore.py - Format Loaders
loaders.py - PII Redaction
pii.py - Chunk Strategies
chunking.py - Chroma Vector Store
Question Answering
- Q&A API
app.py - Retrieval Patterns
retrieval.py - Evidence Assembly
context.py - Cited Answer
app.py
External Services
- Embedding Model
- Chat Model
Clients and Inspection
- HTTP Demo Client
agent.py - Store Inspector
inspect_vectorstore.py
Other components
- Questioner
Connections 14
- Index Builder → Format Loaders: loads units
build_vectorstore.py - Source Documents → Format Loaders: provides files
README.md - Index Builder → PII Redaction: redacts text
build_vectorstore.py - Index Builder → Chunk Strategies: chunks text
build_vectorstore.py - Index Builder → Embedding Model: embeds chunks
build_vectorstore.py - Index Builder → Chroma Vector Store: writes vectors
build_vectorstore.py - Questioner → Q&A API: asks question
README.md - HTTP Demo Client → Q&A API: sends HTTP requests
agent.py - Q&A API → Retrieval Patterns: requests matches
app.py - Retrieval Patterns → Chroma Vector Store: searches vectors
retrieval.py - Q&A API → Evidence Assembly: assembles evidence
app.py - Q&A API → Chat Model: requests answer
app.py - Q&A API → Cited Answer: returns citations
app.py - Store Inspector → Chroma Vector Store: inspects contents
inspect_vectorstore.py