Agents for Humans 2026 · Good Neighbor Agents

Escola Lendária
Community Agent

A proactive school coordination agent that notices silent learner risk before a learner has to ask for help. This public demo executes a live, read-only scan against privacy-minimized learner progress and returns only anonymous risk signals.

{ } Verify JSON Judge fast path ↗

Live source · 2026-11-13T10:30:08.718Z

Learners scanned95

Watchlist95

Human attention95

High priority95

7+ day inactivity95

Repeated failure3

Silent-risk monitor — live evidence

Temporary aliases only. Raw learner IDs are used server-side solely to read records and are never returned to this public page or JSON verifier.

Learner	Risk	Evidence	Route
Learner 01	90	36 days inactive	Human review
Learner 02	90	104 days inactive	Human review
Learner 03	90	106 days inactive	Human review
Learner 04	90	104 days inactive	Human review
Learner 05	90	105 days inactive	Human review
Learner 06	90	103 days inactive	Human review
Learner 07	90	104 days inactive	Human review
Learner 08	90	103 days inactive	Human review
Learner 09	90	103 days inactive	Human review
Learner 10	90	105 days inactive	Human review

Judge proof — four things to verify

The public page proves the read-only live source. The repository then makes the full agent claims reproducible instead of asking judges to trust the pitch.

Live community signal

Refresh the page or inspect JSON to see a fresh privacy-minimized scan.

Deterministic priority

The full agent ranks human work outside the LLM from urgency, active risk and waiting time.

Human authority

Payment, access, discipline, enrollment, deletion, safeguarding and medical/legal decisions cannot be model-executed.

Reproducible evidence

CI runs security, lint, tests and a live-HTTP judge smoke test covering the end-to-end safety path.

Verification report ↗Smoke test source ↗

Safety invariants

✓ Read-only public demoNo write endpoint and no mutation of learner state.

✓ No learner identity outputNames, raw IDs, contacts and PINs are never rendered.

✓ Consequential actions stay humanPayments, access, discipline and safeguarding are not AI-executable tools.

✓ Deterministic model-note guardrailModel-authored notes are independently validated before persistence.

What the full agent adds

Strands + Bedrock reasoning after deterministic safety.
The repository adds persistent conditions, deduplication, clearing, deterministic attention ranking, human follow-ups, audit trails, privacy-safe community briefing, five constrained Strands tools, Amazon Bedrock reasoning and an AgentCore runtime adapter.

Source code ↗End-to-end video ↗

Architecture at a glance

Real Escola Lendária learner progress ↓ read-only / privacy-minimized Autonomous monitor + persistent state ↓ Deterministic guardrail policy + attention ranking ↓ safe contextual cases only Strands Agents SDK + Amazon Bedrock ↓ five constrained tools Human attention queue → human resolution → audit trail ↳ Amazon Bedrock AgentCore runtime adapter

Why this is a Good Neighbor Agent

A request-only chatbot sees the learner who speaks. Community Agent is designed to notice the learner who goes quiet. The beneficiary is the school community as a whole: one agent reduces repetitive monitoring work while surfacing only cases where human judgment is useful.

Measurement boundary: this demo shows operational risk evidence and coordination behavior; it does not claim causal improvement in grades, retention or completion.