eDaftar – AI document drafting for Sindh government offices

Official correspondence in Sindh government offices was handled on paper. Locating a file could take days, drafting routine letters was slow and repetitive, and there was no reliable audit trail of who approved what. Records are sensitive, so any digital system had to enforce strict access control rather than simply speed things up.

A records and correspondence platform with three AI-assisted parts. An OCR pipeline reads uploaded documents and extracts structured data. An LLM assistant drafts the first version of official letters, which staff edit and approve rather than publish directly. A multi-tier e-approval workflow tracks each file in real time with tamper-proof audit logs, and role-based access control across three permission tiers protects sensitive records. Built with Next.js, Node.js and PostgreSQL over REST APIs. My role, as part of a team at Sukkur IBA University: the OCR integration, the AI drafting assistant and the approval workflow, plus training the government staff who use it. Design lesson: a narrow, reliable AI step with a human approval gate earned user trust faster than a broader, less predictable feature would have.

Document processing time reduced by about 70%. File retrieval went from days to seconds. In daily use by 100+ civil servants, with every approval traceable in the audit log.