Responsible AI & backend systems

Trailhead Customer Support API

How can a support assistant help without inventing policy or business authority?

Completed · Tested · 2026

Context

It demonstrates how model-driven conversation can stay bounded by tested authorization, business rules, citations, and clear production-safety limits.

Design problem

How can a support assistant help without inventing policy or business authority?

Approach

Keep conversation and tool selection in the model while Python owns authorization, calculations, policy retrieval, and escalation rules.

Validation

Tests mock Claude; CI runs the test suite and Ruff. Responses expose citations, tool use, and escalation decisions.

Result

Structured support responses distinguish a requested action from a completed action and expose supporting sources.

Limitations

Read-only fictional business demo. Customer email is a demo authentication stand-in; no refund or cancellation is actually performed.

What I would improve next

Production use would require signed identity, permission checks, idempotency, audit logs, and explicit action-success responses.

My role

System architecture, backend implementation, policy grounding, and validation

Repository & documentation

Documentation checked Sep 3, 2026. The summaries describe the repository’s validation approach; they do not claim new hardware or deployment results.

Related field notes

All engineering work

On this device

Settings

Welcome in.

Your theme, audience, name, goal, and saved items stay on this device.

Theme
Privacy & Data

Optional session identity

Sign in

Sign in with Google to personalize your greeting during this browser session. Sign-in does not unlock private content or create a permanent account. Anonymous traffic and performance metrics are measured separately through Cloudflare Web Analytics.

Google is contacted only after you open this account panel.

Review privacy details

Search the fieldbook

Field note