Pipeline Architecture

Multi-agent RAG pipeline with intelligent document processing

Processing Pipeline

Document Ingestion

Stage 1

MineRU-style document parsing and extraction

PDF ParserDOCX ExtractorImage Handler

Multi-OCR Pipeline

Stage 2

DeepSeek OCR 2 • Gemini 3 Flash • Mistral OCR

Table DetectionEquation ExtractionLayout Analysis

Intelligent Chunking

Stage 3

Chonkie-inspired semantic-aware chunking

Heading DetectionTable PreservationOverlap Control

Knowledge Graph

Stage 4

Neo4j graph database for document relationships

Entity ExtractionRelationship MappingGraph Indexing

Multi-Agent RAG

Stage 5

n8n orchestrated search with CRAG & PageIndex

Context DetectionCorrective RAGCitation Ranking

Search & Retrieval

Stage 6

Perplexity sonar-pro with streaming results

Semantic SearchCode SearchWeb Enrichment

Service Connections

Perplexity (Search)

AI-powered search engine

Groq (LLM)

Fast AI inference

Neo4j (Graph)

Knowledge graph database

n8n (Orchestration)

Multi-agent workflows

Firecrawl (Web)

Web scraping & enrichment

GitHub (Code)

Code search & repos

Technology Stack

Perplexity sonar-pro

LLM

Groq (Llama 3.3)

Graph DB

Neo4j 5

Orchestration

n8n Workflows

OCR

DeepSeek / Gemini / Mistral

Chunking

Semantic (Chonkie)

Doc Parse

MineRU Pipeline

RAG

CRAG + PageIndex