Case study / learning systems

Coach Training ANN

Purpose: help instructional designers train AI Design Studio coaches to give students course-accurate feedback. Teams set the materials the coach is allowed to rely on and review sample responses for factual accuracy, source fidelity, teaching value, and clarity. Those reviews help catch unsupported or off-course answers before they reach students.

A.J. Merlino / Coach training layer
Portfolio simulation / no live model

Evidence before response

Teach the coach
what to trust.

Course material has an order of authority. This training layer keeps that order intact while helping a coach find the strongest eligible evidence and screen its draft before it reaches a learner.

Staff-reviewed training records only

System intent

Make feedback accountable to the course.

The ANN is a supporting quality layer, not the coach itself. It does not generate answers, choose authority, or learn from learners.

01 / CURATE

Set the authority order

Instructional designers define the sources a coach may use, from active course material to approved references.

02 / REVIEW

Train with staff judgment

Approved reviews teach the system how to rank eligible evidence and recognize a response that needs another pass.

03 / RELEASE

Freeze the candidate

Evaluated weights can be approved for use. Student interactions remain outside the training set.

Candidate review

Screen a coach draft

Learner question“What should I do first when framing the problem?”

Draft grounded in active courseCoach / v0.4

Start by naming the situation you want to understand, then describe the people affected and the decision the work needs to inform. Keep the first pass specific enough to test, but open enough to revise once you gather evidence.

Course lesson: Framing a design inquiry

The evidence ranker only reorders passages already eligible under the selected source policy.

Factual accuracy
0.92
Source fidelity
0.97
Teaching value
0.88
Clarity
0.94