pg.Purvesh Gosai
All work
CASE STUDY / AGENTIC AI / MULTI-AGENTProduction

Multi-Agent Logistics Platform

Four production agents across connected coordinator and vendor applications for equipment logistics.

LogisticsEnergy & Utilities
Azure AI FoundryMulti-agent orchestrationHuman-in-the-loopEnterprise integrationsFigma

Four agents are live in production across two connected enterprise applications, giving business users one conversational interface in place of multi-screen lookups.

MY ROLE

Designed the multi-agent architecture across both applications.

PROJECT STATUS

Production

FOCUS

Agentic AI / Multi-Agent · Enterprise AI

01 / BUSINESS CHALLENGE

Start with the problem.

Coordinating ground-protection equipment logistics across internal coordinators and external vendors meant working through manual, fragmented processes spanning allocation, work orders, transfers, rentals, and returns. Business users had to look up vendor, purchase order, inventory, and materials data across multiple screens.

02 / SOLUTION APPROACH

A practical path forward.

Two connected enterprise applications—a coordinator-facing logistics/tracking app and a vendor-facing delivery portal—with a multi-agent architecture designed across both, giving business users one conversational interface in place of multi-screen lookups.

My contribution
  • Designed the multi-agent architecture across both applications.
  • Prototyped both application UIs in Figma before handoff to a software development team, who built the production applications.
  • Built the agents in Azure AI Foundry.
  • Worked directly with business teams to learn their current process, then built and refined the agents around how they work and supported them through adoption.
03 / ARCHITECTURE

How the pieces connect.

Four agents live · illustrative connections

Specialist work. Shared context.

3 / 04
Multi-Agent Logistics PlatformClick a component

Four agents are live in production across the vendor and coordinator applications, promoted through DEV → TEST → UAT → Production. Connections illustrate application responsibility rather than disclose internal orchestration. Fulfillment Planning is planned for release.View case study: Multi-Agent Logistics Platform

Explore the Multi-Agent view
04 / IMPLEMENTATION

From design to workflow.

Vendor-facing app · Two agents live in production

Vendor-facing delivery portal

Order Intake & Validation Agent and Inventory & Availability Agent are live in production. Fulfillment Planning Agent is planned, not live.

Coordinator-facing app · Two agents live in production

Coordinator-facing logistics/tracking app

Demand Intake / Project Readiness Agent and Allocation Agent are live in production.

01

Architecture across both applications

Designed the multi-agent architecture across both applications.

02

UI prototyping and software handoff

Prototyped both application UIs in Figma before handoff to a software development team, who built the production applications.

03

Agent development

Built the agents in Azure AI Foundry.

04

Business collaboration and adoption

Worked directly with business teams to learn their current process, then built and refined the agents around how they work and supported them through adoption.

THE AGENTS

Live agents and planned work.

Vendor-facing app

Order Intake & Validation Agent

LIVE in production
Vendor-facing app

Inventory & Availability Agent

LIVE in production
Coordinator-facing app

Demand Intake / Project Readiness Agent

LIVE in production
Coordinator-facing app

Allocation Agent

LIVE in production
Vendor-facing app

Fulfillment Planning Agent

Planned
Storm response

Storm-response agent

Production status not claimed

Purpose

Surfaces material availability and regional shortages.

Human checkpoint

Employee confirmation before acting on any recommendation.

Release path

DEVTESTUATProduction

All four live agents were promoted through DEV → TEST → UAT → Production, with testing and business-user acceptance before go-live.

07.5 / TRUST & GUARDRAILS

How the system fails safely.

Human confirmation

Human confirmation before consequential actions.

Validation before a write

Validation before any write.

Governed enterprise integration

Agents connect to enterprise systems through governed, standard paths.

08 / OUTCOMES

What the work demonstrates.

  • Four agents live in production.
  • One conversational interface replacing multi-screen lookups.

Production status applies to the four named live agents. Fulfillment Planning Agent is planned, not live. Storm-response work is described without a production claim. The production applications were built by a software development team after UI prototype handoff.

NEXT CASE STUDY

Enterprise Knowledge AI Assistant

05 /A CONVERSATION AWAY

Let’s build
what’s next.

Interested in enterprise AI, agentic systems,
or intelligent automation? Let’s connect.

LinkedIn