How we build

What we can stand behind.

This page separates practices in place across PineWoodsAI, controls designed per engagement, and things we do not claim.

In place

Across PineWoodsAI today.

Relational Intelligence Fabric

Studio products sit on an orchestration layer we call the RIF. It combines specialized roles, relevant context, and more than one model provider, then evaluates the combined result. It is not a single chatbot and it is not a product we sell from this site.

Model providers

Language, speech, and related work run through third-party providers. The current Transparency page names the model providers used by Athena Prime. Client work uses the providers agreed in that engagement—not a hidden default. Named client work is collected on the Work page.

Website conversations

Ask Paul is the AI version of Paul Petro on this site. It answers from our public pages. A person on our team can join that conversation when a visitor asks for a human.

Known limitations

The assistant can be incomplete or wrong. It does not have access to private client systems, medical records, or anything beyond this site unless a visitor types it into the conversation.

Incident and support contact

Write arch@pinewoodsai.com. We do not publish a contractual response-time commitment on this site.

Per engagement

Designed for the work, then written down.

Data minimization and flows

Before we access client information, we agree what is required. Implementation documents the relevant data flows, providers, access boundaries, retention, and human-review points for that work.

Identity and access

Who can see what is defined per system. We do not claim a single company-wide access standard that covers every client environment.

Human review

Where a recommendation or generated output can affect a decision, the engagement says who reviews it and when. That is a design choice for the work, not a slogan.

Evaluation and monitoring

Criteria are set before launch and should match the job: accuracy, completeness, consistency, latency, cost, adoption, escalation, or an operating outcome. We do not publish a generic scorecard.

Training-data policy

Whether a provider may use engagement data to train its models is specified in the engagement and in that provider’s terms. We do not claim a blanket “we never train” rule that we cannot enforce on every subprocessor.

Not claimed

We will not imply what is not in place.

  • Certifications, audit reports, or SOC-style attestations
  • Published service levels or uptime guarantees
  • A contractual control set that applies before an engagement is signed
  • That the RIF is available as a standalone product on this site

Product-specific detail for Athena Prime remains on Transparency. Named client work is on Work.

Start here

Bring us something worth seeing clearly.

A product idea. A stubborn process. A difficult decision. A part of the organization that works—but not as well as it should. We'll start there.