Government
Document intelligence for a government entity
AI reads and routes incoming documents, with human review on every decision.
- Duration
- 8 months
- Team
- 11 people
- Services
- AI, Data
The challenge
What was in the way.
The entity received thousands of letters, forms and attachments each month in Arabic and English. Staff read each one, typed key details into a case system and forwarded it to the right department. Backlogs built up, and some requests waited weeks before anyone looked at them. Any automation had to run on government approved hosting, and a person had to stay accountable for every decision.
What we built
The solution.
A document pipeline that reads scanned and digital documents, extracts key fields, suggests a category and department, and places each item in a review queue. Staff confirm or correct each suggestion, and their corrections improve the system over time.
- Document intake
- OCR and extraction
- Classification model
- Human review queue
- Case management system
How we worked
The decision that made the difference.
A pilot on one document type, then four more types added in stages. Every release was checked by the entity’s security team before going live.
Results
What changed.
- 60%
Less time per document
- 82%
Suggestions accepted without change
- 45 days
Backlog cleared
Documents reach the right team the same day, and staff spend their time on decisions instead of data entry.
Technologies used
- Python
- Azure AI Document Intelligence
- PyTorch
- PostgreSQL
- Government cloud