Environment
Simulated OMS data
The prototype uses mock order data so the workflow can be explored without exposing PII.
Product Artifact
OMS ChatGPT App
A prototype ChatGPT App that enables users to interact with Order Management Systems through a conversational interface, handling workflows like order lookup and cancellation with built-in guardrails.
Enterprise WorkflowConversational AIAI AgentsPrototype
- Environment
- Simulated OMS data
- Workflow scope
- Lookup + cancellation
- Design priority
- Trust and control
The prototype uses mock order data so the workflow can be explored without exposing PII.
Focused on two concrete OMS jobs where guardrails and user confirmation matter.
Reasoning, confirmation, and clear user control were treated as first-class product requirements.
Designed to explore how AI agents can safely operate within enterprise workflows.
Ask about this page
Get a grounded read on responsibility, evidence, impact, or what to read next.
Workflow scope
Lookup + cancellation
Focused on two concrete OMS jobs where guardrails and user confirmation matter.
Design priority
Trust and control
Reasoning, confirmation, and clear user control were treated as first-class product requirements.
Product overview
What it is
A tangible workflow artifact, not just a concept statement.
A conversational AI interface for Order Management workflows, designed to make common support and operations tasks faster, clearer, and safer to execute.
Problem framing
Why I built it
OMS workflows are often fragmented across multiple tools, slow to execute, and difficult to automate safely because of PII constraints.
At the same time, early AI efforts were typically either standalone experiences or embedded inside existing workflows. There was not yet a clear pattern for AI interacting directly with internal systems through conversation.
This prototype explored that pattern.
Workflow scope
What it does
Focused on a narrow slice of OMS work where trust and control matter.
- Retrieve order details through natural language
- Execute order cancellations with confirmation and safeguards
- Provide transparent reasoning behind actions
- Maintain user control throughout the interaction
Interaction design
How it works
Built as a custom ChatGPT App, the prototype uses mock OMS data to simulate real workflows without exposing PII. It was designed to live inside the existing AI platform environment employees were already using.
The goal was not just to simulate functionality. It was to design for trust, clarity, and safe execution.
Prototype walkthrough
Live product states
Actual screens from the prototype showing how lookup, confirmation, and completed actions are handled across the workflow.
Public web prototype
The standalone web version makes the core interaction pattern visible outside the ChatGPT shell while preserving the same OMS workflow structure and guardrail logic.
This is the live experience linked from the page.
Order lookup inside ChatGPT
Natural-language lookup returns a structured order view with status, shipping details, totals, and expandable items.
Cancellation with guardrails
The confirmation state makes the risk explicit, requires a deliberate phrase, and keeps the user in control before submission.
Completed state after action
After confirmation, the assistant shows the updated order state clearly instead of hiding the result behind a generic success message.
Strategic takeaway
Why it matters
This prototype introduced a new interaction model: AI acting directly on internal systems through conversation.
It helped shift thinking from AI as a standalone tool toward AI as an active participant in real workflows. Framed carefully, it represented an early example of a new pattern for conversational interaction with internal systems.
Follow-on work
From prototype to real system work
The prototype created alignment to move forward and explore production viability.
- Evaluated SSO integration for secure user access
- Defined PII-safe interaction patterns
- Built an internal MCP server to support agent workflows
- Explored a production-oriented architecture using Java for team alignment and a Python wrapper for the Agents SDK
Builder signal
What this proves
- AI can operate within enterprise constraints when designed intentionally
- Prototypes can unlock faster alignment than strategy documents alone
- Product leaders can de-risk platform investments through hands-on builds
Scope note
Notes on the prototype
- Uses simulated data to respect the sensitivity of OMS systems
- Represents a functional concept rather than a production deployment
- Focused on validating interaction patterns, not replacing existing systems
Contact
Interested in the product logic, guardrails, or enterprise workflow angle?
This page is structured to make the prototype easy to discuss with recruiters, builders, and teams thinking about AI in real operating environments.