robayedl/documind

How robayedl/documind works

DocuMind is a multi-tenant document intelligence product: users upload PDFs or DOCX files, which are parsed and indexed asynchronously, then ask questions answered with hybrid retrieval and a revisable multi-agent RAG workflow. The graph also preserves documented MCP access, usage tracking, caching, and privacy/auth capabilities. Some relationships come from README architecture statements and the source index; unsampled implementations are represented at their known source paths, without inferring their internal call details.

Main entry points: main.py FastAPI service · researcher.py Research and tool use · retrieval.py Hybrid retrieval and HyDE · tasks.py Ingestion task

Components 26

Web client

Chat interface web/app/chat/ChatClient.tsx
API client web/lib/api.ts

API and tenancy

FastAPI service app/main.py
Clerk authentication app/auth.py
PII redaction app/redact.py
Semantic cache rag/cache.py
PostgreSQL and pgvector app/db.py

Document ingestion

Ingestion task worker/tasks.py
Celery queue worker/celery_app.py
Document parsing and indexing rag/ingest.py
Chunk contextualization rag/contextualize.py
Uploaded-file storage app/storage.py

RAG answering

Agent supervisor graph rag/agents/graph.py
Research and tool use rag/agents/researcher.py
Hybrid retrieval and HyDE rag/chains/retrieval.py
Cross-encoder reranking rag/chains/rerank.py
Chunk and vector store rag/store.py
Answer synthesis rag/agents/synthesizer.py
Answer critique rag/agents/critic.py
Search and calculator tools rag/tools
Gemini and embeddings rag/llm.py

Integrations and usage

MCP document tools mcp_server/server.py
API-key validation mcp_server/auth.py
Token and cost usage rag/usage.py

Other components

User
LLM and search providers

Connections 28

User → Chat interface: asks questions
Chat interface → API client: sends requests web/app/chat/ChatClient.tsx
API client → FastAPI service: calls API
FastAPI service → Clerk authentication: validates users app/main.py
FastAPI service → PII redaction: redacts queries app/main.py
FastAPI service → Semantic cache: checks responses app/main.py
FastAPI service → Agent supervisor graph: runs agent graph app/main.py
Agent supervisor graph → Research and tool use: dispatches research
Research and tool use → Hybrid retrieval and HyDE: retrieves passages rag/agents/researcher.py
Hybrid retrieval and HyDE → Chunk and vector store: searches chunks rag/chains/retrieval.py
Hybrid retrieval and HyDE → Cross-encoder reranking: reranks candidates rag/chains/retrieval.py
Research and tool use → Search and calculator tools: calls selected tools rag/agents/researcher.py
Research and tool use → Answer synthesis: provides research
Answer synthesis → Answer critique: submits draft
Answer critique → Answer synthesis: requests revision
FastAPI service → Ingestion task: queues ingestion app/main.py
Ingestion task → Celery queue: runs as task worker/tasks.py
Ingestion task → Document parsing and indexing: indexes documents worker/tasks.py
Document parsing and indexing → Chunk contextualization: contextualizes chunks rag/ingest.py
Document parsing and indexing → Chunk and vector store: writes indexed chunks rag/ingest.py
FastAPI service → Uploaded-file storage: saves uploads app/main.py
FastAPI service → PostgreSQL and pgvector: persists metadata app/main.py
Chunk and vector store → PostgreSQL and pgvector: reads and writes vectors rag/store.py
MCP document tools → API-key validation: validates API keys mcp_server/server.py
MCP document tools → Hybrid retrieval and HyDE: searches documents mcp_server/server.py
Agent supervisor graph → Token and cost usage: tracks usage rag/agents/graph.py
Gemini and embeddings → LLM and search providers: invokes Gemini rag/llm.py
Search and calculator tools → LLM and search providers: uses web search

Drawn by GitDiagram on November 13, 2026 from the repository's file tree, README and source files. Read as Markdown ·

How robayedl/documind works

Main entry points: main.py FastAPI service · researcher.py Research and tool use · retrieval.py Hybrid retrieval and HyDE · tasks.py Ingestion task

Connections 28

User → Chat interface: asks questions
Chat interface → API client: sends requests web/app/chat/ChatClient.tsx
API client → FastAPI service: calls API
FastAPI service → Clerk authentication: validates users app/main.py
FastAPI service → PII redaction: redacts queries app/main.py
FastAPI service → Semantic cache: checks responses app/main.py
FastAPI service → Agent supervisor graph: runs agent graph app/main.py
Agent supervisor graph → Research and tool use: dispatches research
Research and tool use → Hybrid retrieval and HyDE: retrieves passages rag/agents/researcher.py
Hybrid retrieval and HyDE → Chunk and vector store: searches chunks rag/chains/retrieval.py
Hybrid retrieval and HyDE → Cross-encoder reranking: reranks candidates rag/chains/retrieval.py
Research and tool use → Search and calculator tools: calls selected tools rag/agents/researcher.py
Research and tool use → Answer synthesis: provides research
Answer synthesis → Answer critique: submits draft
Answer critique → Answer synthesis: requests revision
FastAPI service → Ingestion task: queues ingestion app/main.py
Ingestion task → Celery queue: runs as task worker/tasks.py
Ingestion task → Document parsing and indexing: indexes documents worker/tasks.py
Document parsing and indexing → Chunk contextualization: contextualizes chunks rag/ingest.py
Document parsing and indexing → Chunk and vector store: writes indexed chunks rag/ingest.py
FastAPI service → Uploaded-file storage: saves uploads app/main.py
FastAPI service → PostgreSQL and pgvector: persists metadata app/main.py
Chunk and vector store → PostgreSQL and pgvector: reads and writes vectors rag/store.py
MCP document tools → API-key validation: validates API keys mcp_server/server.py
MCP document tools → Hybrid retrieval and HyDE: searches documents mcp_server/server.py
Agent supervisor graph → Token and cost usage: tracks usage rag/agents/graph.py
Gemini and embeddings → LLM and search providers: invokes Gemini rag/llm.py
Search and calculator tools → LLM and search providers: uses web search

Drawn by GitDiagram on November 13, 2026 from the repository's file tree, README and source files. Read as Markdown ·