Guided product walkthrough
BeforeBell demo
Three synthetic school-coverage scenarios show what BeforeBell automates, where it stops for human judgment, and how it safely recovers when the first plan fails.
Centerpiece scenario
Daniel Reed — where automation stops on purpose.
BeforeBell covers P2 and P3 through routine policy, detects that P5 crosses a protected-planning boundary, interrupts the live agent, and resumes only after an administrator selects a permitted option.
Resets only the fixed synthetic Scenario B demo state.
01
Routine coverage
P2 and P3 are safely assigned.
02
Judgment boundary
P5 cannot be safely automated.
03
Human decision
Administrator selects one permitted option.
04
Trusted fulfillment
Approved path executes and resolves 3/3.
Autonomous
Routine success
Sarah Miller
A single safe candidate can cover the entire Math absence. No administrator judgment is required.
Centerpiece
Human judgment
Daniel Reed
Routine Science coverage succeeds for P2/P3, but P5 crosses a protected policy boundary.
Resilient
Safe fallback
Olivia Chen
The preferred candidate declines, so BeforeBell preserves the decline, replans, and safely moves to the next eligible candidate.
What this demo proves
Deterministic policy
Eligibility, conflicts, protected planning and assignment safety stay outside the model.
Real agent orchestration
Strands coordinates the workflow through Amazon Bedrock and deployed AgentCore.
Human-in-the-loop
The agent cannot choose or fabricate an exception. Administrator judgment arrives through a real interrupt.
Authoritative evidence
Assignments, human decisions, and workflow events are persisted in DynamoDB.
Design principle
Safe routine decisions happen automatically.
Judgment stays human. BeforeBell is intentionally narrow: absence coordination, safe coverage, exception handling and resolution — nothing more.
Synthetic demo data only. Riverside Community School and all people shown here are fictional training entities.