Inspiration

Public sector organizations across Vietnam and Southeast Asia still rely heavily on manual, multi-step workflows to classify, route, review, and respond to administrative documents. This process averages 5-7 days per document, with limited visibility into document status and inefficient coordination across departments. Documents pass through 6 manual steps — intake, registration, distribution, review, consultation, and response — creating bottlenecks, misrouting risks, and duplicated effort.

We asked ourselves: How might we leverage AI to transform this broken workflow while respecting strict security and confidentiality requirements across 4 classification levels (Unclassified → Top Secret)?

What it does

DocRoute AI is an AI-powered document intelligence platform that automates the entire 6-step administrative document workflow:

  • Automated Ingestion: Processes both physical (scanned) and digital documents using OCR and intelligent parsing
  • Intelligent Classification: Qwen AI automatically detects document type, subject matter, and security classification level
  • Smart Routing: AI-driven department identification and automatic routing to the correct processing unit
  • AI Summarization: Extracts key information, generates concise summaries, and identifies action items
  • Cross-Department Collaboration: Structured workflow enabling parallel review instead of sequential processing
  • Real-time Tracking: Full lifecycle visibility with status dashboards and audit logging
  • Multi-level Security: Role-based access control across Unclassified, Confidential, Secret, and Top Secret levels

How we built it

We built DocRoute AI on a production-grade BPM (Business Process Management) platform enhanced with Qwen AI at its core:

  • Qwen AI Integration: Used Qwen models for document classification, Vietnamese language NLP, summarization, and entity extraction
  • OCR Pipeline: Integrated OCR for processing scanned physical documents with varying quality levels
  • Workflow Engine: Built automated routing rules that map document types to departments based on AI classification output
  • Security Layer: Implemented role-based access control with 4 classification levels and comprehensive audit logging
  • Dashboard: Real-time tracking interface showing document lifecycle status across all departments

Challenges we ran into

  • Handling variable quality scanned documents required robust pre-processing before OCR and AI analysis
  • Vietnamese language document processing required careful prompt engineering with Qwen to handle administrative terminology
  • Balancing automation speed with security constraints — ensuring classified documents never bypass access controls
  • Designing a system that works in hybrid paper-digital environments where document ID numbers are not fully standardized
  • Cross-department workflow coordination with multiple disconnected legacy storage systems

Accomplishments that we're proud of

  • Reduced document processing time from 5-7 days to minutes for standard documents
  • Built a working prototype that handles all 6 steps of the administrative workflow automatically
  • Achieved accurate Vietnamese language document classification and summarization using Qwen AI
  • Implemented proper multi-level security with audit trails suitable for government use
  • Created a system that bridges the gap between physical and digital document workflows

What we learned

  • Qwen AI excels at multilingual document understanding, especially with Vietnamese administrative text
  • Government document workflows have universal pain points that are highly automatable with the right AI approach
  • Security and accessibility must be designed in from day one, not bolted on later
  • The biggest efficiency gains come from eliminating duplicated cross-department review, not just speeding up individual steps

What's next for DocRoute AI

  • Pilot deployment with Vietnamese government agencies for real-world validation
  • Expand language support for other Southeast Asian administrative contexts
  • Add predictive analytics for document processing time estimation and workload balancing
  • Integrate with existing government e-document systems and digital signature platforms
  • Scale the platform to handle higher document volumes with enterprise-grade infrastructure
Share this project:

Updates