Selected work
Supply chain operationsApplication family

From a delayed material to an approval-ready recovery plan.

This application family was built around a common supply chain problem: the first signal of a disruption may arrive in an inbox, while the operational effect is spread across supplier, material, production and customer records. The system uses several bounded agents to assemble that picture without giving software authority to make a commitment.

ApplicationSupply Chain Operations Agent
EnvironmentWindows workflow + browser voice cockpit
Agent model3 bounded roles
Decision ownerAccountable person
The operating situation

The problem was larger than a single prompt.

01

Purchasing requests and supplier updates entered through a shared inbox.

02

The consequence of a delay was not visible in one place.

03

Staff had to trace the exposure before they could compare recovery actions.

04

Any supplier, production or customer commitment still needed accountable approval.

What the team needed

A useful system had to fit the work and its limits.

  • Keep Outlook and Excel in the workflow rather than forcing an immediate platform replacement.
  • Let a manager ask a spoken question and receive a traceable operational briefing.
  • Apply known routing rules before asking AI to interpret an unresolved request.
  • Prepare recovery options as an editable action packet for human review.
System design

Each agent has a separate job.

A realtime voice agent handles the spoken briefing. Routing and recovery agents collect evidence, trace impact and prepare options so a person can make the operational decision.

A1

Routing agent

Reads incoming purchasing work, applies deterministic company rules first and sends only unresolved cases for constrained AI review.

A2

Impact-tracing agent

Connects the supplier or material issue with relevant production and customer records, keeping the evidence attached to the briefing.

A3

AI voice briefing agent

Lets the manager ask follow-up questions in a browser-based voice cockpit, with clear disclosure that the conversation is with AI.

A4

Recovery agent

Compares available actions and compiles an approval-ready packet. It can prepare the choice, but it cannot make the commitment.

Working sequence

One traceable path from input to review.

  1. 01Ask the AI voice agent about a delay or exposed commitment
  2. 02Trace the effect through materials, production and customers
  3. 03Compare recovery scenarios and purchasing actions
  4. 04Approve, edit or reject the action packet
Platform and stack

Technology chosen around the operating environment.

OpenAI Realtime APIOpenAI Responses APIWebRTC voicePowerShell 5.1Outlook + Excel COMWindows DPAPI
Desktop workflow

PowerShell 5.1 with Outlook and Excel COM on managed Windows computers.

AI services

OpenAI Realtime API for voice and the OpenAI Responses API for bounded reasoning tasks.

Voice layer

WebRTC conversation in a browser interface with written evidence and action details.

Local protection

Windows DPAPI for machine-level protection of sensitive local configuration.

Deployment

A browser-based voice cockpit deployed on Render, separate from the restricted desktop workflow.

What the business can measure

Evidence for the next decision, not a promised result.

Time to trace exposure

From the first disruption signal to a usable view of affected work.

Request assignment

How long routine and exceptional purchasing requests wait for an owner.

Recovery preparation

Time and rework required to assemble options for approval.

Control quality

Actions that reach review with their evidence, assumptions and accountable owner intact.

Evidence boundary

This page describes a working application family and its control model. Client names, operating data and identifying details are intentionally omitted. No financial savings or production outcomes are claimed here.

Have a similar problem?

Start with the work, the constraint and the decision owner.

We can map the process, decide which agents would help and define a first release that is small enough to evaluate.

Discuss the problem