An agentic supply chain management system, built from scratch.

Supply chain decisions were being made in inboxes, spreadsheets and the heads of a small number of people who had been there long enough to know what "normal" looked like.

That works until volume rises or one of those people leaves. The failure mode is not dramatic — nothing crashes. Decisions just get slower, less consistent, and harder to explain after the fact.

Narrative drafted from the brief — replace with the engagement detail

1. A

   The bottleneck was the routine decisions, not the exceptional ones.

2. B

   Nothing recorded why a call had been made, only what it was.

3. C

   The proposed dashboard would have described the problem without changing it.

An agentic system that handles the routine decisions and escalates the ones that genuinely need a person, with the reasoning recorded either way.

• Decision engine

  Java services taking the routine calls against the rules the business already ran on, with the reasoning written to Postgres beside the decision.

• Escalation path

  Anything outside the rules goes to a person, in a JavaScript front end that carries the context with it rather than making it be re-gathered.

• Infrastructure and integrations

  Provisioned on AWS with Terraform, and wired into the systems already in use so nothing had to be re-keyed to make it work.

• Handover

  Their team was in the pull requests during the build, not briefed at the end of it.