ERNIE & PaddlePaddle Hackathon 2025

Multi-Agent Document Intelligence

Powered by ERNIE LLM and PaddleOCR-VL. Multiple AI agents working together to understand, analyze, and answer questions about your documents.

Activate Neural Core Explore Architecture

Neural Agent Network

Watch data flow through the cognitive pipeline

IN Document OCR PaddleOCR-VL CORE Coordinator CAMEL-AI ANA Analysis SUM Summary Q&A Dialog OUT Intelligence
Idle
C

Coordinator Agent

The neural core that orchestrates all agents. Decomposes tasks, manages data flow, and implements CAMEL-AI RolePlaying for deep analysis.

Task Planning RolePlaying Aggregation
O

OCR Agent

Vision-language perception layer. Extracts text from documents using PaddleOCR-VL with layout understanding and structure preservation.

PaddleOCR-VL Layout Analysis Markdown
A

Analysis Agent

Deep understanding layer. Extracts entities, classifies documents, and identifies key information using ERNIE's comprehension capabilities.

ERNIE 4.5 NER Classification
S

Summary Agent

Synthesis layer. Distills complex documents into key points, generates structured outlines, and creates concise summaries.

ERNIE 4.5 Abstractive Key Points
Q

QA Agent

Interactive layer. Enables natural language dialog with documents, supports multi-turn conversations with citation capabilities.

ERNIE 4.5 Multi-turn Citations

Live Neural Processing

Feed a document into the neural network and watch it think

Input Signal

Waiting for input signal...

The neural network will show real-time agent activity here

Technical Deep Dive

ERNIE

Baidu ERNIE LLM

Powered by Baidu's ERNIE large language model:

  • Document analysis and entity extraction
  • Abstractive summarization
  • Context-aware question answering
  • Multi-turn dialog management
PaddleOCR

PaddleOCR-VL

PaddlePaddle's vision-language OCR model:

  • High-accuracy text extraction
  • Layout and structure analysis
  • Table recognition
  • Multi-language support
CAMEL-AI

Multi-Agent Framework

Agent collaboration using CAMEL-AI patterns:

  • Task decomposition and planning
  • RolePlaying for deep analysis
  • Coordinator orchestration
  • Result aggregation