Flagship systems

Applied AI · Deterministic document systems

DocAgent: AI Document Intelligence Pipeline

A controlled AI pipeline that preserves document structure instead of flattening everything into text.

3 input modes
System architecture01 — 05
01Confluence & Word
02OOXML Extractor
03Copilot Agent
04Human Review
05DOCX Renderer

My scope

The boundary I owned.

I designed a deterministic Python shell around Copilot for ingestion, OOXML extraction, asset tracking, review, and native rendering.

  • Python
  • GitHub Copilot
  • OOXML
  • python-docx
  • Confluence
  • Word COM

Engineering evidence

  1. Accepted Confluence pages, Word documents, and pasted content through one normalized ingestion path.
  2. Used stable image markers and direct OOXML extraction to preserve links, tables, images, and procedural order.
  3. Constrained the probabilistic stage to semantic restructuring while deterministic code handled conversion, assets, rendering, and validation.
OutcomeInspectable transformations with human approval at the final boundary.