Live product leadership case study

Companion Intelligence

Making consequential decisions easier to trust, challenge and defend by connecting evidence, reasoning, uncertainty and human accountability.

ProductDecision-assurance demonstrator
My roleProduct founder & delivery lead
StatusLive public MVP
DataEntirely synthetic
01 · The problem

Consequential decisions can become difficult to explain - and easy to distort.

Important decisions are often assembled across fragmented documents, scoring workbooks, meeting discussions and executive papers. When the decision is challenged, an organisation may struggle to establish what was known, what was excluded, how conclusions changed and who owned the final judgement.

The product opportunity was not simply to generate a more persuasive answer. It was to preserve and expose the complete path from source evidence to accountable human decision.

02 · The product response

A governed decision-assurance layer

Companion Intelligence connects the evidence chain and the reasoning chain without collapsing them into one opaque output.

SourceEvidence itemEvidence packageControl status
ObservationFindingAssessmentJudgementDecision

The demonstrator uses a synthetic NZ$85 million managed-technology procurement to show how omitted evidence, unsupported score changes, source dependency, undeclared conflicts and non-normalised pricing affect decision readiness.

03 · What I led

End-to-end product leadership

I acted across the complete product lifecycle:

  • Product strategy and MVP scope
  • Governance and operating principles
  • Evidence and information architecture
  • Explainability requirements
  • User journeys and interaction design
  • Synthetic case and test design
  • Acceptance and confidence testing
  • Repository, deployment and release governance
04 · Key product decisions

Judgement mattered more than feature volume

Prove one complete vertical slice.One bounded case was taken from evidence through reasoning, explanation and human intervention rather than building a broad but shallow platform.
Separate evidence from reasoning.What was collected and preserved remains distinct from what people or systems concluded from it.
Distinguish Fact, Inference and Human Judgement.The interface prevents interpretation from being presented as established fact.
Prefer honest gaps over plausible invention.When the record does not explain a score change, the product says so and recommends re-moderation.
Keep accountability human.The system may recommend an action; the accountable human owns the decision.
Do not overstate the technology.The MVP uses structured data, deterministic controls and curated question paths rather than pretending to be an unrestricted production AI platform.
05 · Explainability by design

A product contract, not an afterthought

Every material answer is structured around:

ConclusionEvidence basisReasoning traceConfidenceUncertaintyLimitationsHuman review

The system must prefer an honest gap over a plausible invention.

The unsupported-question pathway and missing-reasoning pathway were tested as first-class product behaviour - not treated as edge cases.

06 · AI-native delivery

AI increased delivery speed; it did not replace product accountability.

Companion Intelligence was created through an AI-native workflow combining human product judgement, systems thinking and governance with AI-assisted research, design, drafting, coding and testing.

I retained ownership of the product problem, scope, architecture, acceptance criteria, trade-offs and release decisions. That distinction is central to how I approach responsible adoption: AI accelerates the work, while accountable humans remain responsible for what is built and claimed.

07 · Outcome

From concept to a live, testable product outcome

7linked product screens
6 + 1explainability and honest-gap tests passed
LiveGitHub-to-Vercel deployment pipeline
SSLcustom-domain release operational

The result is a recruiter-accessible product that demonstrates product strategy, responsible-AI governance, systems architecture, interaction design and delivery discipline in one coherent artefact.

08 · Product boundaries

What the MVP proves - and what it does not claim

Included

  • One complete synthetic decision case
  • Structured evidence and reasoning records
  • Deterministic control findings
  • Explainability and failure-state behaviour
  • Human decision-readiness recommendation
  • Public read-only deployment

Not claimed

  • Production authentication or permissions
  • Customer document ingestion
  • Unrestricted natural-language retrieval
  • Persistent enterprise workflow
  • Multi-tenant architecture
  • Production assurance certification
09 · Role relevance

The capability bridge

This product demonstrates capability relevant to AI product leadership, data and AI transformation, responsible AI, AI governance and assurance, decision intelligence, cyber transformation and technology delivery leadership.

I am not positioning myself as a machine-learning engineer or data scientist. My strength is turning emerging capability into products that are useful, governable, explainable and operationally credible.

See the product behaviour

Open the live Companion Intelligence demonstrator.

Follow the executive journey, inspect the evidence chain, challenge the reasoning and test how the product handles an unsupported question.