Inspiration
Procurement teams spend a significant amount of time on repetitive work: finding suppliers, comparing quotes, checking supplier risk, reviewing documents, negotiating terms, preparing purchase orders, and following up on deliveries.
We wanted to turn this workflow from a collection of manual tasks into an autonomous process. The idea behind SupplyGuardian AI is simple: let an AI agent handle the operational workload while keeping humans in control of important decisions.
This led us to build an autonomous procurement agent that can reason over procurement requests, use external tools, coordinate specialized agents, and escalate only the decisions that genuinely require human judgment.
What it does
SupplyGuardian AI is an autonomous procurement and supply-chain agent for enterprise teams.
A procurement request can trigger an end-to-end workflow:
- Understand the procurement requirement and create an execution plan.
- Discover and evaluate potential suppliers.
- Compare pricing, quality, delivery terms, and supplier history.
- Assess financial, compliance, and supply-chain risks.
- Analyze supplier responses and identify the best options.
- Prepare negotiation recommendations and supplier communications.
- Coordinate approvals and human review when required.
- Generate procurement outputs such as supplier shortlists, comparison reports, risk assessments, negotiation summaries, and purchase orders.
- Track fulfillment and surface exceptions or emerging risks.
The goal is not to replace procurement professionals. Instead, SupplyGuardian acts as an AI procurement teammate that handles repetitive operational work and keeps humans in the loop for high-impact decisions.
How we built it
SupplyGuardian AI is built around an agentic architecture using the Strands Agents SDK.
The system is organized into specialized agents responsible for different stages of the procurement workflow:
- Planner Agent — understands the request and creates an execution plan.
- Research Agent — discovers and evaluates suppliers.
- Analysis Agent — compares suppliers, pricing, quality, and risk.
- Negotiation Agent — prepares negotiation strategies and supplier communications.
- Execution Agent — coordinates procurement actions, approvals, purchase orders, and tracking.
These agents share relevant context and can use tools to interact with supplier data, market information, communication systems, procurement systems, and other external services.
AWS services provide the supporting infrastructure, including Amazon Bedrock for foundation-model capabilities, Amazon S3 for document storage, structured databases for procurement data, and AWS Lambda/CloudWatch for supporting services and monitoring.
We also designed the system around a human-in-the-loop model. The agent can autonomously progress through routine tasks, but sensitive actions such as approving a supplier, accepting a negotiation outcome, or committing a purchase can be routed to a human for approval.
Challenges we ran into
One of the biggest challenges was deciding where autonomy should end and human control should begin. A procurement agent cannot simply execute every action it recommends. We therefore designed explicit approval and escalation points for high-impact decisions.
Another challenge was coordinating multiple specialized agents without losing context. Supplier information, procurement requirements, previous decisions, risk findings, and negotiation context need to remain available throughout the workflow.
We also had to think carefully about tool reliability and failure handling. Real procurement workflows involve incomplete documents, inconsistent supplier information, unavailable data, and unexpected responses. The system therefore needs to verify information and surface exceptions instead of blindly continuing.
Finally, we focused on making the agent's actions understandable to the user. Procurement teams need to know why a supplier was recommended, what risks were detected, and what the agent intends to do next.
Accomplishments that we're proud of
We are proud of turning a traditionally fragmented procurement workflow into a single agentic experience.
Instead of building another procurement dashboard or chatbot, we designed SupplyGuardian as an active worker that can plan, research, analyze, negotiate, execute, and escalate.
We are particularly proud of the human-in-the-loop design. The objective is not maximum autonomy at any cost; it is useful autonomy with controlled execution.
We also built the architecture around specialized Strands agents and tool use so that the system can be extended with additional procurement capabilities without redesigning the entire workflow.
What we learned
Building SupplyGuardian taught us that effective AI agents need more than a powerful language model.
They need:
- Clear task boundaries
- Reliable tools
- Persistent context
- Structured outputs
- Verification steps
- Failure handling
- Human approval mechanisms
- A well-defined execution loop
We also learned that the best agent experiences are not necessarily the ones that automate everything. The strongest design is often one where the agent handles the repetitive work autonomously and brings the human into the workflow exactly when their judgment adds the most value.
What's next for SupplyGuardian AI
Our next goal is to move SupplyGuardian from a procurement assistant into a broader autonomous supply-chain operations platform.
Future capabilities could include:
- Continuous supplier risk monitoring
- Predictive supply disruption detection
- Automatic alternative-supplier discovery
- Dynamic price and market analysis
- Multi-round autonomous negotiation
- Contract intelligence
- Inventory-aware procurement
- Supplier performance prediction
- Automated purchase-order and invoice workflows
- Deeper ERP and procurement-system integrations
- Learning from historical procurement decisions
Ultimately, we want SupplyGuardian to become a trusted AI procurement teammate that continuously protects an organization's supply chain while keeping humans firmly in control of consequential decisions.
Built With
- agentic-ai
- ai-agents
- amazon-bedrock
- amazon-dynamodb
- amazon-rds-relational-database-service
- amazon-web-services
- api
- autonomous-agents
- aws-cloudwatch
- aws-lambda
- business-process-automation
- enterprise-ai
- generative-ai
- human-in-the-loop
- llm
- multi-agent-systems
- procurement-automation
- python
- risk-analysis
- strands-agents-sdk
- supplier-risk-management
- supply-chain-management
- tool-calling
- uipath-maestro
- workflow-automation
Log in or sign up for Devpost to join the conversation.