nitishkumarpidikiti457/enterprise_rag_and_agentic_ai

How nitishkumarpidikiti457/enterprise_rag_and_agentic_ai works

This platform answers enterprise questions using document retrieval, read-only SQL, or both, and returns grounded answers with citations. The user-facing flow is client → authenticated FastAPI endpoint → single-shot RAG or the LangGraph agent → retrieval and/or SQL → routed model response. A separate ingestion flow turns uploaded or corpus documents into indexed chunks. The README documents streaming, conversation memory, model fallbacks, caching, and metrics; unsampled modules are shown at their documented boundaries, with uncertain wiring omitted.

Main entry points: main.py FastAPI service · pipeline.py Ingestion pipeline · retriever.py Hybrid retrieval · graph.py LangGraph workflow

Components 24

API service

FastAPI service src/app/api/main.py
JWT and limits src/app/api/auth.py
Chat UI src/app/api/static/index.html
Upload ingestion src/app/api/main.py

Document ingestion

Ingestion pipeline src/app/ingestion/pipeline.py
Document loaders src/app/ingestion/loaders.py
Semantic chunking src/app/ingestion/chunking.py
Embedding src/app/ingestion/embed.py

Retrieval and data

Hybrid retrieval src/app/retrieval/retriever.py
Cross-encoder reranker src/app/retrieval/rerank.py
Chroma vector store src/app/retrieval/vectorstore.py
SQLite business data data/sample.db

Agent and answers

Single-shot RAG src/app/rag.py
LangGraph workflow src/app/agent/graph.py
Agent tools src/app/agent/tools.py
Thread memory src/app/agent/graph.py

Models and operations

LLM router src/app/llm/router.py
LLM providers src/app/llm/providers.py
Prompt templates src/app/llm/prompts/templates.yaml
Response cache src/app/llm/cache.py
Metrics and logs src/app/observability/logging.py

Other components

Client
Redis cache
Model services

Connections 31

Client → FastAPI service: sends requests
FastAPI service → JWT and limits: authenticates and limits
FastAPI service → Chat UI: serves UI
FastAPI service → Upload ingestion: routes uploads
Upload ingestion → Ingestion pipeline: ingests files
Ingestion pipeline → Document loaders: loads documents
Ingestion pipeline → Semantic chunking: chunks documents
Semantic chunking → Embedding: embeds sentences
Ingestion pipeline → Embedding: embeds chunks
Ingestion pipeline → Chroma vector store: upserts chunks
Ingestion pipeline → Hybrid retrieval: invalidates BM25
FastAPI service → Single-shot RAG: answers queries
FastAPI service → LangGraph workflow: runs or streams chat
LangGraph workflow → Agent tools: dispatches tools
Agent tools → Hybrid retrieval: searches documents
Agent tools → SQLite business data: queries read-only
LangGraph workflow → LLM router: classifies and synthesizes
LangGraph workflow → Thread memory: persists thread state
Single-shot RAG → Hybrid retrieval: retrieves passages
Single-shot RAG → LLM router: generates answers
Hybrid retrieval → Embedding: embeds query
Hybrid retrieval → Chroma vector store: searches dense index
Hybrid retrieval → Cross-encoder reranker: reranks candidates
LLM router → LLM providers: selects and calls
LLM router → Response cache: reads and writes
Response cache → Redis cache: uses optional backend
LLM providers → Model services: calls model APIs
Single-shot RAG → Prompt templates: renders prompts
LangGraph workflow → Prompt templates: renders prompts
LLM router → Metrics and logs: records calls and cost
FastAPI service → Metrics and logs: records service metrics

More Python diagrams: make-all/tuya-local · impact-i/reflutter · idea-research/dino · powerdns-admin/powerdns-admin · openenv/openenv

Drawn by GitDiagram on September 26, 2026 from the repository's file tree, README and source files. Read as Markdown ·