Employee
Alyssa starts as a Customer Support Lead working through a configured customer incident.
Explore a candidate workflow for training people and generating AI skill configuration.
This uncalibrated public candidate demonstrates one coaching workflow using configured source excerpts and deterministic rules. No live model, organizational certification, or access-control system is connected.
See the modelTry the coaching labWorking product slice
Coach one real workflow from baseline to reusable skill.
Alyssa is a Customer Support Lead. This uncalibrated candidate scores her exact learner text with a deterministic rubric configured against the demo sources below.
Browser-local demo state
- 1Baseline
- 2Coach
- 3Practice
- 4Reassess
- 5Package
Baseline · Northstar Health export incident
Customer escalation summary
Write a customer-ready escalation summary using only the approved incident facts. A reviewer should be able to identify impact, evidence boundaries, ownership, workaround, and update timing.
- 37 analysts cannot export quarter-end reports.
- The failure began after a 9:15 AM MT SSO configuration change and has lasted 4 hours.
- The timing is confirmed; root cause is not confirmed.
- Maya Chen, Support Incident Lead, owns coordination.
- Safe workaround: audited CSV export from the admin console.
- Next customer update: 2:30 PM MT, even if root cause is still unknown.
- Maya Chen must review and approve the draft before external delivery.
Submit baseline artifact
This first draft becomes a browser-local assessment snapshot when submitted.
Use fictional demo data only; do not enter customer, employee, or confidential information.
178 characters · 40 minimum
Product vision
Train people first. Let reviewed evidence inform a portable workflow skill, then let connected systems decide access.
Role-aware learning flow
Evidence before review
Candidate skill output
Concept model
One candidate path from baseline practice to reviewable workflow configuration.
Coaching
Deterministic rules compare submitted text with an uncalibrated rubric derived from synthetic source excerpts.
Candidate review
Scores are demo signals tied to submitted evidence—not certification, manager approval, or an access decision.
Candidate package
A portable JSON configuration shows instructions and review requirements; no live model adapter is included.
Candidate output
A portable configuration for a connected runtime to adapt.
The demo assembles instructions, examples, source references, blocked-behavior guidance, and review requirements. Its JSON download is not an executable integration, permission set, certification record, or access grant; a separately secured runtime must adapt and enforce it.
Candidate configurationsupport.customer-escalation-summary
Configured role contextConfigured source examplesSource referencesReview requirementsBlocked-behavior guidanceEvidence references
ChatGPT
Potential adapter target for instructions, examples, and review rules. No connector is included.
Claude
Potential adapter target for project instructions and templates. No connector is included.
Copilot
Potential adapter target for required checks and escalation policy. No connector is included.
Internal assistant
Requires separate identity, policy enforcement, source ingestion, reviewer workflow, and audit controls.
Governance
Reviewed candidate evidence can inform access decisions in a connected control system.
Configured source rules
Configured excerpts support this demo; production requires approved ingestion and version control.
Demo evidence
Browser-local artifacts show the concept; they are not immutable organizational records.
Deployment conditions
Skill instructions guide behavior; connected identity and policy systems enforce access.