Case Study

AI Document Classification Agent

Automating manual file routing with a lightweight, AI-powered agent to reduce repetitive work and eliminate misfiled documents.

AI Automation Workflow Optimization LLM Agents Node.js Operations
Manual → Auto
Routing
Zero-Touch
File Archiving
High
Confidence Threshold
↓ Misfiles
Error Reduction
!

The Problem

Scanned documents required employees to manually open each file, read its contents, mentally categorize the document type and client, manually rename the file, and then drag it into the correct directory. This workflow was repetitive, slow, and highly prone to misfiling errors.

✓

The Solution

A lightweight AI-powered agent. It automatically reads, decides, and routes scanned files. It features a transparent History Log window and an explicit Human-in-the-Loop toggle for user control. By relying on structured LLM reasoning and CSV client mapping, it eliminates manual triage while ensuring reliability.

Workflow Transformation

From tedious manual clicks to autonomous routing.

Manual Routing

  • ×
    Visually inspect each scanned PDF
  • ×
    Mentally identify the client and document type
  • ×
    Manually rename the file (prone to typos)
  • ×
    Drag-and-drop to specific deeply-nested folders
  • =
    Slow, repetitive, with high risk of misfiled documents

AI Agent Workflow

  • ✓
    System detects new scan and extracts text via OCR
  • ✓
    LLM evaluates text against a CSV client database
  • ✓
    Returns structured JSON (client, type, confidence, and reasoning)
  • ✓
    App automatically routes the file and logs the decision history
  • =
    Consistent, fast, scalable file organization

Data Pipeline Architecture

Product Strategy

Strategic Takeaways

MVP Over Enterprise Bloat

It is tempting to build a huge enterprise platform, but operational value often comes from small, robust tools. A well-written Node.js application deployed on a Windows Server was the most practical choice.

Trust Through Transparency

AI isn't perfect. Implementing an explicit Human-in-the-Loop toggle and a History Log window built trust. Users knew they could audit the process or intervene if the system was uncertain.

Explainable AI (XAI) as a Feature

The process improved dramatically when the LLM was prompted to return its reasoning for a decision. This "Chain of Thought" increased accuracy and populated the History Logs, allowing users to understand exactly why a file was routed a certain way.

My Role

Acted as a Technology Consultant and Developer. I observed the manual workflow, conceptualized the automation agent, wrote the Node.js application, engineered the LLM prompts, and deployed the solution to their local server.

Tech Consultant / Developer

Tools & Concepts

Node.js
LLM Prompting
Chain of Thought
Audit Logging
OCR / Text Extraction
File Automation
Human-in-the-Loop