chandancoding/production-agentic-rag

How chandancoding/production-agentic-rag works

This is a provider-agnostic agentic RAG library with CLI and FastAPI interfaces. Its main workflow loads and chunks documents into a hybrid index, then checks a question, plans sub-questions, retrieves and synthesizes grounded answers, and returns citations and a trace. Evaluation, alternate LangGraph execution, and optional integrations are separate capabilities. Edges follow shown calls or README claims; some branches are documented but their implementation was not sampled.

Main entry points: pipeline.py RAG pipeline · rag_agent.py Agent loop · hybrid.py Hybrid retriever · cli.py CLI

Components 21

Interfaces

CLI src/production_agentic_rag/cli.py
HTTP service api/main.py
RAG pipeline src/production_agentic_rag/pipeline.py

Ingestion

File loaders src/production_agentic_rag/ingestion/loaders.py
Chunk strategies src/production_agentic_rag/ingestion/chunking.py

Retrieval

Hybrid retriever src/production_agentic_rag/retrieval/hybrid.py
BM25 search src/production_agentic_rag/retrieval/bm25.py
Vector store src/production_agentic_rag/retrieval/vectorstore.py
Embedding providers src/production_agentic_rag/providers/embeddings.py

Agent and safeguards

Agent loop src/production_agentic_rag/agents/rag_agent.py
Query planner src/production_agentic_rag/agents/planner.py
Retrieval tool src/production_agentic_rag/agents/tools.py
LLM providers src/production_agentic_rag/providers/llm.py
Guardrails src/production_agentic_rag/guardrails.py
LangGraph engine src/production_agentic_rag/graph.py

Evaluation and telemetry

Evaluation harness src/production_agentic_rag/evaluation/harness.py
Evaluation metrics src/production_agentic_rag/evaluation/metrics.py
Trace accounting src/production_agentic_rag/observability.py
Optional integrations src/production_agentic_rag/integrations.py

Other components

User
Document corpus

Connections 28

User → CLI: asks or evaluates README.md
User → HTTP service: sends request README.md
CLI → File loaders: loads documents src/production_agentic_rag/cli.py
CLI → RAG pipeline: runs query src/production_agentic_rag/cli.py
Document corpus → File loaders: provides files README.md
File loaders → RAG pipeline: supplies documents src/production_agentic_rag/cli.py
RAG pipeline → Chunk strategies: chunks text src/production_agentic_rag/pipeline.py
RAG pipeline → Hybrid retriever: indexes chunks src/production_agentic_rag/pipeline.py
RAG pipeline → Guardrails: checks and redacts src/production_agentic_rag/pipeline.py
RAG pipeline → Agent loop: runs selected engine src/production_agentic_rag/pipeline.py
RAG pipeline → Embedding providers: builds embedder src/production_agentic_rag/pipeline.py
RAG pipeline → LLM providers: builds provider src/production_agentic_rag/pipeline.py
RAG pipeline → Trace accounting: records query trace src/production_agentic_rag/pipeline.py
RAG pipeline → LangGraph engine: selects when configured src/production_agentic_rag/pipeline.py
Agent loop → Query planner: decomposes query src/production_agentic_rag/agents/rag_agent.py
Agent loop → Retrieval tool: calls retrieval src/production_agentic_rag/agents/rag_agent.py
Agent loop → LLM providers: synthesizes answer src/production_agentic_rag/agents/rag_agent.py
Agent loop → Guardrails: scores grounding src/production_agentic_rag/agents/rag_agent.py
Agent loop → Trace accounting: records usage spans src/production_agentic_rag/agents/rag_agent.py
Retrieval tool → Hybrid retriever: retrieves chunks src/production_agentic_rag/agents/tools.py
Hybrid retriever → BM25 search: searches lexically src/production_agentic_rag/retrieval/hybrid.py
Hybrid retriever → Vector store: searches vectors src/production_agentic_rag/retrieval/hybrid.py
Vector store → Embedding providers: embeds chunks src/production_agentic_rag/retrieval/vectorstore.py
HTTP service → RAG pipeline: serves RAG requests README.md
CLI → Evaluation harness: runs eval src/production_agentic_rag/cli.py
Evaluation harness → Evaluation metrics: computes measures src/production_agentic_rag/evaluation/harness.py
Optional integrations → Hybrid retriever: wraps retrieval tool README.md
Trace accounting → Optional integrations: exports telemetry README.md

More Python diagrams: upstox/upstox-python · waltsims/k-wave-python · youngskkim/crn · adobe-research/nolima · alouetter/alavette-form

Drawn by GitDiagram on September 30, 2026 from the repository's file tree, README and source files. Read as Markdown ·