Research Demo Paper

DocMaster: A Hierarchical Structure-Aware System for Document Analysis

We present DocMaster, a hierarchical structure-aware document analysis system. DocMaster parses documents into hierarchical document trees preserving original layouts and constructs a structure-aware semantic index that enables accurate document filtering and in-depth analysis.

Paper Codebase Frontend Demo Learn More

Note: The online demo is a static frontend preview and is not connected to the backend. To run the complete system, visit the DocMaster codebase.

Capabilities

Key Features

DocMaster combines structural and semantic analysis to deliver efficient, accurate document analysis.

Document Tree Search

Builds a hierarchical tree from structural elements. The LLM traverses top-down, pruning irrelevant branches to minimise token usage.

Hyperedge Search

Extracts cross-chunk semantic relationships as hyperedges. FAISS-based retrieval finds the most relevant hyperedges for relation-aware filtering.

Combined Strategy

Aggregates evidence from both strategies. Fusing structural and relational signals achieves higher precision and recall than either alone.

Token-Efficient Design

Tree-traversal prunes irrelevant subtrees early, significantly reducing LLM token consumption compared to naive full-document retrieval.

RAG Q&A Integration

After filtering, perform retrieval-augmented Q&A over matched documents using the same indexed embeddings for seamless document intelligence.

Multi-Document Collections

Upload an entire collection of documents. The system processes, indexes, and evaluates the filter condition across all documents in a single query.

How It Works

System at a Glance

From document ingestion to semantic filtering and RAG-powered analysis.

Workflow pipeline of DocMaster

Workflow Pipeline

End-to-end flow from PDF upload through parsing, indexing, filtering, to RAG Q&A output.

Index construction pipeline

Index Construction

Document tree building, PC-KMeans clustering, and hyperedge extraction pipeline.

Demo system layout

Demo System Layout

Two-column interface with document management on the left and filtering/chat on the right.

Explore More

Architecture

Progressive filtering pipeline and system component stack.

Explore →

Video

Watch a full walkthrough of the system in action.

Watch →

User Guide

Step-by-step instructions for uploading, filtering, and querying.

Read →

Examples

Interactive walkthroughs of filtering and Q&A scenarios.

View →