hrishikeshchandi/semantic-drift-rag

How hrishikeshchandi/semantic-drift-rag works

Semantic Drift RAG is a document-grounded question-answering service. Users upload documents, which are loaded, chunked, embedded, indexed for hybrid retrieval, and used to derive corpus drift statistics. For a query, the API checks document scope before invoking a LangGraph retrieval, response, and evaluation loop; out-of-scope queries are refused and near-boundary queries request clarification. The README documents capabilities beyond the sampled implementation, including citations, multi-session use, and multiple ingestion formats; those links are shown where documented rather than inferred from unsampled code.

Main entry points: main.py FastAPI server · vectorstore.py Hybrid vector store · nodes.py RAG operations · files.py File management API

Components 14

API and interface

Web interface frontend/script.js
FastAPI server main.py
File management API routers/files.py

Document ingestion

Document processor document_ingestion/processor.py
Hybrid vector store vectorstore/vectorstore.py
User index and sessions

Scope and sessions

Drift detector drift_detector/detector.py
Session drift state drift_detector/detector.py

Grounded answering

RAG graph graph_builder/builder.py
Pipeline state state/rag_state.py
RAG operations nodes/nodes.py
LLM configuration llm/llm.py

Other components

User
Model providers

Connections 20

User → Web interface: uses
Web interface → FastAPI server: submits queries
User → FastAPI server: uploads and queries README.md
FastAPI server → File management API: mounts routes main.py
File management API → Document processor: loads and chunks routers/files.py
File management API → Hybrid vector store: builds index routers/files.py
Document processor → Hybrid vector store: provides chunks README.md
Hybrid vector store → User index and sessions: persists index README.md
FastAPI server → Hybrid vector store: loads retriever main.py
FastAPI server → Drift detector: analyzes query main.py
Drift detector → Session drift state: reads and updates drift_detector/detector.py
Drift detector → User index and sessions: reads corpus stats drift_detector/detector.py
FastAPI server → RAG graph: runs answering graph main.py
RAG graph → RAG operations: dispatches steps graph_builder/builder.py
RAG graph → Pipeline state: invokes with state graph_builder/builder.py
RAG operations → Pipeline state: updates nodes/nodes.py
RAG operations → Hybrid vector store: retrieves passages README.md
RAG operations → LLM configuration: calls models nodes/nodes.py
LLM configuration → Model providers: configures llm/llm.py
Hybrid vector store → Model providers: uses embeddings README.md

Drawn by GitDiagram on November 15, 2026 from the repository's file tree, README and source files. Read as Markdown ·