NeighborAid Queue

Inspiration

Small food banks often triage requests through spreadsheets and messages. Volunteers repeatedly check consent, urgency, requested items, and stock. NeighborAid Queue removes that routine work while keeping consequential decisions with a human coordinator.

What it does

The agent turns minimized request data into inventory-safe allocation plans and local audit records. Routine requests become AUTO_READY; shortages or same-day requests become HUMAN_REVIEW; missing consent or direct personal data is BLOCKED.

How we built it

The Python prototype uses the Strands Agents SDK with three typed tools: request validation, inventory planning that never permits negative stock, and minimal audit creation. A batch orchestrator invokes each capability through Strands' direct tool interface.

Challenges and accomplishments

The challenge was balancing automation with privacy and operational safety. We used synthetic data, deterministic inventory reconciliation, strict data minimization, and an explicit human gate. The repository includes an MIT license, architecture documentation, a reproducible CLI demo, and automated tests for state transitions, stock reconciliation, audit output, and PII blocking.

What's next

The current build is local and credential-free so nonprofits can evaluate it at no cost. The same tool contracts can later run with a model provider and Amazon Bedrock AgentCore while preserving the human gate. No eligibility decision, outbound message, purchase, or real-world reservation is automated.

Built With

• agents
• python
• sdk
• strands