GenAI Procurement

Designing and building a multi-agent AI system for enterprise procurement

Role: UX / Service Designer · GenAI Engineer
Focus: AI Product Design · Service Design · GenAI Engineering · AI Governance
Tech: React · Flask · Azure AI Foundry · LangChain · RAG · Vector Search

Outcome

The project resulted in a working multi-agent AI prototype that demonstrates how procurement workflows could move from:

Manual → Assisted → Intelligent

without removing humans from critical decisions.

More importantly, the project helped me develop an approach to AI product design that connects UX, service design, AI engineering and AI governance.

What I learned

AI changes the role of the designer.

Designing an AI product isn't only about designing screens.

It requires understanding:

People → Service → AI behaviour → Data → Engineering → Governance

My role increasingly sits at the intersection of these disciplines.

The takeaway

Don't automate the workflow simply because AI can.

Redesign the workflow around what AI is uniquely good at—while keeping people in control.

The approach

I approached the problem as a service transformation, rather than simply adding a chatbot.

01 — Understand the service

Mapped the end-to-end procurement journey and identified high-friction activities where AI could provide meaningful value.

02 — Identify AI opportunities

Four high-value agent opportunities emerged:

• RFP Agent
  Creates compliant RFPs from procurement requirements.

• Supplier Q&A Agent
  Handles repetitive supplier questions using organisational knowledge.

• Evaluation Agent
  Helps compare suppliers against multiple criteria.

• Contract Agent
  Assists with contract drafting and compliance.

03 — Design the human–AI workflow

The agents were designed as collaborators, not autonomous replacements.

The system provides:

AI recommendation → Evidence → User review → Human decision

This keeps people in control of high-impact decisions.

From prototype to working system

I didn't stop at the interface.

I built a working technical prototype to understand the practical constraints of the AI experience.

Architecture

React
Frontend

↓

Flask
Backend / API

↓

Azure AI Foundry
AI orchestration

↓

Specialised agents + RAG

↓

Vector Search / enterprise data

↓

Structured AI output

This allowed me to validate both the UX and the underlying AI workflow.

AI engineering

The prototype explored:

• Multi-agent orchestration

• RAG and vector search

• Structured Prompts

• Structured outputs

• Agent memory

• Tool and knowledge retrieval

• Prompt and response validation

• Error and fallback handling

• Agent evaluation

• Observability and feedback loops

Responsible AI by design

Procurement is a high-impact environment, so governance was considered alongside the product design. Key considerations included:

• Transparency — users should understand AI-generated recommendations.

• Human oversight — important decisions remain with people.

• Data privacy — supplier and procurement information must be protected.

• Auditability — AI decisions and actions should be traceable.

• Bias & fairness — supplier recommendations should be evaluated for unintended bias.