Agent Pageant - Agentic Document Classifier

Agentic Automation Agent Pageant
, , , , , , ,
1

Agent Pageant submission type

agent-for-everyone

Full name

Aleksandra Kwiecień

Team name

SolAI

Team members

@Aleksandra_Kwiecien

How many agents do you use

one-agent

Industry category in which use case would best fit in (Select up to 2 industries)

logistic

Complexity level

intermediate

Summary (abstract)

From the chaotic docks of logistics to the silent corridors of law firms and finance departments – document chaos is everywhere. Our Agentic Document Classifier is the first truly ‘plug-and-play’ solution for PDF separation. No more manual sorting, no more language barriers, and no more rigid templates. Whether it’s an invoice in Hungarian or a contract in Polish, our AI Agent identifies, categorizes, and splits files with surgical precision.

Detailed problem statement

The Challenge: In global operations, especially in sectors like logistics, finance, and HR, organizations are overwhelmed by “bundled” PDFs. Instead of receiving one file per document, they get a single 100-page scan containing a random, unsorted mix of invoices, transport notes (CMRs), and orders.

Why Traditional Classifiers Fail?

  • Boundary Blindness: Legacy RPA classifiers ask “What is this file?” but cannot answer “Where does the Invoice end and the CMR begin?” within a single stream.
  • The Language Barrier: Standard tools require manual, brittle rule-building for every language (e.g., Hungarian, Polish, Greek), which is impossible to maintain at scale.
  • Rigid Schemas: Traditional ML models are “frozen” after training; adding a new document type requires developer intervention and model retraining.
  • Manual Bottlenecks: This lack of intelligence forces employees to spend hours manually scrolling and “splitting” PDFs before any actual automation can start.

The Solution Gap: There is a critical need for an “Agentic” approach that doesn’t just match keywords but reasons through the document content. We need a system that can:

  • Detect logical boundaries between documents automatically.
  • Understand context regardless of language or layout without pre-defined keyword lists.
  • Operate in the background, processing high volumes without locking user sessions or robot licenses.

Detailed solution

To solve the challenge of unstructured document bundles, I built a hybrid automation that combines the reasoning power of Generative AI Agents with the reliability of RPA processes. The solution is designed as a “plug-and-play” classification engine that requires zero hard-coding for new document types.

  1. The Orchestration Layer (UiPath Apps & RPA Process)
    A specialized UiPath App allows users to upload files and define classification rules on the fly.

  2. The “Agentic” Brain (Dual-Mode Reasoning)
    The core of the solution is a UiPath Agent with a System Prompt that switches between two distinct methodologies based on user input:

  • General Mode: If no specific rules are provided, the Agent uses its internal LLM knowledge to identify documents in any language (e.g., Hungarian, Polish). It assigns standard, professional English categories based on the semantic content of each page.
  • Context Mode: For specific business processes, users can provide your own Document Types with descriptions. The Agent then acts as a strict mapping logic, matching page content only against the provided descriptions and returning exact, case-sensitive keys.
  1. Integrated RPA Toolset
    The Agent uses a sequence of specialized RPA tools to handle the heavy lifting:
  • StorageBucket_DownloadDocument: Securely retrieves the file from a storage bucket for processing.
  • PDF_GetContent: Extracts text and structure from the PDF using OCR.
  • PDF_SplitDocuments: Once the Agent determines the logical boundaries (e.g., “Page 1-3 is an Invoice, Page 4 is a CMR”), this tool physically splits the original PDF into separate, categorized files.

Narrated video link (sample: https://bit.ly/4pvuNEL)

Expected impact of this automation

Whether used by an operations clerk via a dedicated App or integrated into a fully autonomous RPA process, this solution delivers a 70-90% reduction in manual document preparation time. It turns document chaos into structured, actionable data, significantly increasing the throughput of the entire automation ecosystem.

UiPath products used (select up to 4 items)

UiPath-agent-builder
UiPath-apps
UiPath-orchestrator
UiPath-studio

Automation Applications

none

Integration with external technologies

none

Agentic solution architecture (file size up to 4 MB)

PDD_AgentPegeant_AgenticDocumentClassifier.docx (934 KB)

Sample inputs and outputs for solution execution

Other resources

1 Like