AI Training Profile

The AI Training Profile is Wisteria’s answer to a problem most AI tools punt on: “What does training-relevant mean for THIS organisation?”

A document about IV-line infections is critical training for a hospital and irrelevant for a courier company. The AI evaluator can’t know that from the document alone — it needs your business context. The AI Training Profile is where you give it that context.

Two layers

The profile has two layers that compose:

When the AI evaluator looks at a candidate file for the Finance department, it sees: company baseline + Finance department profile. For the Operations department, it sees: company baseline + Operations department profile.

What the company baseline contains

The super_admin sets the baseline at Settings → AI Training. The fields are deliberately open-ended:

• What does your organisation do? — the elevator pitch
• Who are your customers / who do you serve? — patients, retail consumers, enterprise clients, etc.
• What kinds of training matter most? — compliance, safety, sales, technical, customer service
• What’s NOT training-relevant? — exclusion list; legal documents, financial reports, marketing brochures, etc.
• Tone and culture notes — anything that should colour the AI’s framing

The form is plain text per field. You write in your own words; Claude reads it the same way a new hire would.

What the department profile adds

A trainer can add department-specific context at the same Settings page (their own department only). Typical content:

• Department mission — what this team is responsible for
• Department-specific training priorities — beyond the company baseline
• Department language — terms of art, acronyms, role names
• Department exclusions — what NOT to surface for this team specifically

The trainer edits their own department’s profile. They can read other departments’ profiles but not edit them.

How the AI uses it

When the watcher evaluates a candidate file for a specific department:

1. The company baseline is injected as system context.
2. The relevant department profile is injected next.
3. The candidate file’s text is appended.
4. Claude is asked to judge whether the file is training-relevant for this department in this organisation.

The result is a per-suggestion confidence score and a written explanation of why it surfaced — which the trainer sees on the “Wisteria noticed” card.

When to update it

• After your first month of scans, look at “Not training” feedback patterns. Add exclusions to the company baseline if a class of files keeps surfacing wrongly.
• When you expand into a new line of business, update the baseline to reflect it.
• When a department’s role changes — new product, new compliance regime, new tooling — update that department’s profile.

The profile isn’t fire-and-forget. It’s a living description of your organisation; the more current it is, the sharper the AI’s judgments.

What happens if the profile is empty?

You can leave fields blank. Claude will fall back to generic reasoning about training relevance — workable, but you’ll get noisier results than if you’d filled in the fields.

If you’re piloting Wisteria and don’t yet want to think about the profile, leave it empty for the first week, then come back and fill it in based on which suggestions felt off.