GraphLens AI System Architecture Diagram

Django REST Framework exposes authenticated endpoints for chat, multipart PDF ingestion, and user workspace file management.
PyMuPDF and LangChain text splitters chunk incoming PDFs into 1000-character segments with 200-character overlap.
All file records and vector embeddings are multi-tenant and strictly isolated by user_id.

LangGraph Supervisor routes incoming prompts across four specialized worker agents using fast heuristics and LLM classification.
Document RAG Agent scopes retrieval to active files in the user namespace, querying ChromaDB or Pinecone.
Web Researcher Agent runs an autonomous ReAct loop with DuckDuckGo Search and Jina Reader (r.jina.ai).
Workspace File Agent executes CRUD operations with automatic vector re-indexing on disk and S3.
Email Agent resolves recipient and attachment details and transmits mail via central SMTP relay.

Pinecone (serverless) / ChromaDB (local) persists 1024-dimensional document embeddings.
AWS S3 Secure Vault and local media storage store raw files under users/{user_id}/ directory prefixes.
Relational database (SQLite/PostgreSQL) tracks UserFile records, timestamps, and Django authentication users.