Overview

Designing a governed production model for AI-assisted UX work

DPMA explores how AI can support serious UX production without collapsing the process into disconnected prompts, weak traceability, and disposable outputs. The system is organized around structured stages, explicit approval, artifact-centered deliverables, and a real implementation backbone.

The Challenge

From fragmented artifacts to governed production

Important UX work is often difficult to recover once a project is complete. In enterprise environments especially, the story behind a project can become scattered across screenshots, partial documents, remembered workflows, and disconnected artifacts. That makes portfolio reconstruction, case study writing, and long-term documentation much harder than they should be.

Most AI workflows don’t solve that problem well. They can generate useful fragments, but they do not naturally preserve staged logic, approval history, artifact truth, or continuity across a project lifecycle. DPMA was designed to address that gap.

What I Designed

A stage-centered workspace with human authority at the core

I designed DPMA as a governed production model rather than a generic AI chat tool. The system separates human authority, orchestration, specialist production work, and operational validation. That structure makes the workflow behave more like a professional production environment and less like a single assistant improvising everything at once. It creates a foundation for stage gating, artifact verification, and clearer accountability across the system.

Core structure

• Boss: final human authority
• Manager: orchestration without direct tool use
• Specialists: stage- and task-based production work
• Operations / validation: separate from creative generation

Reverse Mode

Reconstructing historical UX work from incomplete evidence

One of the most important pivots in the project was shaping the MVP around Reverse Mode. Instead of focusing only on generating new outputs, I designed the MVP to reconstruct historical UX work from partial evidence such as screenshots, archived pages, remembered workflows, and incomplete documentation. That made the system much more relevant to a real professional problem: rebuilding project history for portfolio and case study use when the original materials are incomplete or unavailable.

Reverse Mode also forced the system to distinguish between what was observed, what was inferred, and what could not be supported confidently. That became one of the project’s strongest principles.

Governance by Design

Approval and artifact truth are structural rules, not conversational assumptions

A defining part of DPMA is that governance is structural, not implied. Stage progression does not happen because a conversation feels finished. It requires explicit review, approval, and confirmation. I also designed the system around artifact truth: a stage is not meaningfully complete unless real deliverables exist. Internal gate files or status markers alone do not count as evidence of work.

That logic shaped the broader artifact strategy, including the distinction between solution artifacts, supporting documentation, and system artifacts.

Interface Direction

Separating conversation from operational stage work

The workspace is stage-centered, not chat-centered. I developed the interface around a persistent Stage Pane and Chat Pane. Chat functions as the communication surface. The stage area functions as the operational workspace for artifacts, logs, stage memory, and governance actions. This keeps important project material from disappearing into chat history and aligns the product more closely with how structured work is actually reviewed.

Workspace principles

• Stage-centered workspace
• Persistent artifact visibility
• Governance actions live with stage artifacts
• Chat supports control, not artifact storage

Implementation Backbone

Translating the system concept into a working repository framework

A major part of the project involved building the repository structure that made DPMA operational. Over multiple months, I developed and refined the framework through Python logic, YAML configuration, Markdown documentation, runtime scaffolding, and tracking artifacts. The repository became the working backbone of the concept, translating workflow ideas into a real implementation structure that included agent definitions, tasks, loaders, gate logic, and runbook support.

Implementation highlights

• Python-based runtime structure
• YAML-defined agents and tasks
• Gate and approval logic
• Runbook and tracking artifacts
• Repository-backed system scaffolding

Outcome

A structured foundation for trustworthy, reviewable UX production

DPMA now exists as both a product concept and a substantial working framework.

Together, these make DPMA a strong portfolio project in its own right, combining UX architecture, workflow design, systems thinking, and implementation-aware product strategy.