Portfolio project: an evaluated document-to-decision workflow

A strong portfolio project shows how a decision is supported, tested and operated. A screenshot of a successful canvas run leaves the important questions unanswered.

Validate the evidence set

The automation lab defines an authored project index.

from automation_cases import portfolio_case

result = portfolio_case()
assert result["required_complete"] is True
assert result["evaluated"] is True
assert result["production_proof"] is False
assert result["reviewer"] == "pending"
assert result["artifacts"]["external_deployment"] is False

The local evidence can demonstrate engineering decisions, but it is not production proof or external approval.

Build the project in layers

Choose a rights-cleared document task with a real decision contract. Create the brief, source/data manifest and manual baseline. Ingest with duplicate/version checks, extract typed fields with page evidence, validate rules and form a proposal. Route low confidence and failures to an owned queue; bind human approval before a stubbed or sandboxed write.

Evaluate field and decision accuracy, unsupported cases, approval integrity, idempotent replay, recovery, latency and cost per accepted task. Include one deliberate regression that the gate rejects. Document threat model, permissions, privacy/retention, license/dependency inventory, runbook, rollback and limitations.

Provide one command or clear sequence that recreates local evidence from fixtures. Demonstrate normal, unsupported and recoverable-failure paths. Label mocks and authored values; never imply deployment, clients or business savings that were not observed.

The Generative & Agentic AI course uses this artifact set to align portfolio evidence with the course’s automation, RAG, agents and evaluation modules.

Submission checklist

Submit repository, README, diagram, schemas, fixtures, tests, result report and five-minute demo. Explain one design trade-off, one adverse result and the human’s remaining responsibility.

Continue learning

This article is part of the Business automation with AI sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the preceding task in Plan a handover for a no-code AI automation.
• Return to the cluster foundation in Design an automation brief before choosing n8n or an agent.

References: NIST AI RMF Generative AI Profile and OWASP Top 10 for LLM and GenAI Applications.