Document Management System#

AI-SAI-Generated Summary H-RHuman Reviewed H-EHuman Edited

One of my final assignments at Transfer Online was to build an internal platform that brought documentation, access control, QA workflows, and AI-assisted tools into one governed system. I designed the application architecture and relational database from scratch, used test-driven development, and established role-based access and audit requirements before adding the later AI capabilities.

Note

Implementation timeline I designed and implemented the underlying application architecture, relational schema, workflows, access controls, and audit requirements before the AI capabilities were introduced. Later additions included semantic search, RAG, chat, and agent actions.

View the complete implementation timeline.

I also added semantic search for every document and documentation site uploaded, as well as set up a chatbot on a RAG pipeline. I gave the chatbot access to specific functionality that would allow users to say something like, “Upload this PDF to this repository,” and the PDF would be uploaded to the database and integrated into the CI/CD. Additionally, I made a lot of the AI workflows for developing manuals more intuitive and easier for someone using the software to create them by filling out a simple form.

The form accepted a site URL, PDF, or other source material with instructions, although I also tested how the workflow handled minimal direction. The agent applied the configured standards and generated a review-ready manual through the existing validation and publishing system. For one roughly 70-page manual, the workflow reduced first-pass production from approximately 1.5 months to under one hour while retaining human review and quality gates. See a slightly redacted example here: redacted manual.

Governance and Access#

User management and role-based access to documentation sites across the organization.
A visible audit trail showing changes and actions in the UI.

Documentation and Collaboration#

A centralized definition library where users could propose terms and use voting to support cross-team agreement.
A Word-like editing experience for existing documentation sites, enabled by the patterned structure of RST and Sphinx.

QA and Operational Workflows#

A QA area with ticketing connected to the codebase and chatbot, including the ability to submit a ticket through conversation.
A graduated troubleshooting and escalation process defined through a structured JSON rule library.