← AI Reference Architectures

[ bounded implementation pattern ]

School accessibility assistant with teacher-controlled review

A limited classroom support pattern for transforming teacher-approved material into an accessibility aid—such as a reading-level draft, caption draft, vocabulary support, or alternative format—without making the assistant an assessment authority.

Boundary

This pattern is not a substitute for individualized education planning, accessibility review, student-data governance, or teacher judgment. It should begin with approved instructional material and an equivalent non-AI path for students who cannot or should not use it.

[ Proposed data flow ]

Trace each handoff before approving the system.

1 · Teacher-approved source

A teacher selects material that is authorized for the activity. The pilot avoids uploading broader student records or unrelated classroom work.

2 · Bounded transformation

A local or approved remote tool creates a draft accessibility aid: for example, a plain-language explanation, caption draft, vocabulary list, or alternate-format candidate.

3 · Teacher review

The teacher checks accuracy, instructional fit, accessibility, and whether the draft changes meaning before it reaches students.

4 · Student access

Students receive the approved aid through the school’s established access path, with an equivalent non-AI route where appropriate.

5 · Feedback & review

The school records access friction, teacher correction, student usefulness, and unexpected data flows before expanding the pilot.

Assumptions to validate

  • The school has determined permitted data, account, retention, and procurement conditions under its local rules and applicable law.
  • The workflow does not delegate grading, discipline, placement, or high-stakes educational decisions to the tool.
  • Teachers retain authority to approve or reject each output used in instruction.
  • The pilot has an accessibility and equitable-access plan for supported devices, assistive technologies, language needs, and alternatives.

Failure and fallback

  • If the tool is unavailable or a draft is unsuitable, the teacher uses the original material and an established accommodation or instructional route.
  • If a student cannot access the tool, the school provides an equivalent non-AI path rather than making the tool a condition of participation.
  • If a data, content, or access concern arises, pause the affected workflow and route it through the school’s existing policy and review process.

Measurement plan

Accessibility usefulness

Use teacher and student feedback plus observed task completion; do not infer accessibility from feature availability alone.

Instructional accuracy

Sample outputs against source material and document corrections, omissions, and meaning-changing errors.

Equity of access

Test actual school and home access conditions, assistive technologies, account availability, and language needs.

Data boundary

Record prompts, uploads, identifiers, logs, accounts, retention, and any third-party processing used by the pilot.

Sources and implementation references

  1. U.S. Department of Education: Protecting Student Privacy

    Points schools to guidance for evaluating how online educational services collect, use, and transmit student information.

  2. U.S. Department of Education: Student Privacy Guidance

    A current repository of student-data privacy guidance and resources.

  3. IES: Equal Access to Education Websites

    An education accessibility reference; a usable interface still requires context-specific testing.

  4. NIST AI Risk Management Framework

    A voluntary framework for managing AI risks in design, development, use, and evaluation.

Plan a boundaryAssess readiness