Senteras
Text us a question Book a call
Industry

AI in Education

Institutional AI that satisfies FERPA before it touches a student record.

Education is the sector where AI adoption has run furthest ahead of governance. Individual staff are already using consumer tools on student work, usually without anyone having checked whether that constitutes a disclosure of an education record. Senteras builds the institutional alternative: capable systems running on infrastructure the institution controls, so the compliance question has a straightforward answer.

Book a free consultation

What makes this hard

  • Staff are already using consumer AI tools on student data, and most institutions cannot say with confidence what has been disclosed
  • FERPA makes an education record disclosed to a vendor a legal event, not a procurement detail
  • Under-13 students bring COPPA obligations that most general-purpose AI tools explicitly disclaim in their terms
  • Budget cycles do not match the pace at which the tooling changes, so procurement decisions age badly

Where AI earns its place

Administrative Automation

Draft the routine correspondence, scheduling and reporting that consumes administrator time, with no student identifiers in the model context where they are not needed.

Hours per week returned to staff

Student Support Triage

Route and draft first responses to the high-volume enquiry categories, escalating anything touching wellbeing or safeguarding to a human immediately and unconditionally.

Faster first response, human judgment preserved

Instructional Material Generation

Generate differentiated versions of existing material against the institution's own curriculum standards, rather than a generic model's guess at grade level.

Differentiated material in minutes

Institutional Knowledge Retrieval

Natural-language search across policy, handbooks and past decisions so staff stop asking the same question through four channels.

One answer, consistently

Figures are drawn from Senteras engagements and are illustrative of typical results. Outcomes vary by data quality, infrastructure and scope.

The rules that shape the build

These are the constraints that decide the architecture, usually before anyone has picked a model. This is general information about the regulatory landscape, not legal advice on your obligations.

FERPA

An education record disclosed to a vendor requires either consent or the school official exception, which in turn requires direct control over the vendor's use of the data. Self-hosting keeps the record inside the institution and the question does not arise.

COPPA

For under-13 students, verifiable parental consent attaches to the collection of personal information. Most general-purpose AI terms of service explicitly disclaim suitability for this age group.

IDEA and Section 504

AI-assisted content used with students on an IEP or 504 plan still has to meet the accommodations in the plan. Generated material is not exempt.

How we approach it

Safeguarding escalations are never model-decided

Any query that touches student wellbeing, self-harm, abuse or safety routes to a named human immediately and is never answered by the model, even where the model would likely answer well. This is a hard routing rule in the system, not a prompt instruction, because a prompt instruction can be talked out of it and a routing rule cannot.

Where this applies

The same core systems, with the differences that matter in each setting.

K-12 districts
COPPA and FERPA both apply; the practical constraint is usually that staff have already adopted tools nobody approved.
Higher education
FERPA applies but COPPA does not, and the research computing group often already has the GPU capacity for self-hosting.
Edtech companies
You inherit your customers' FERPA obligations through your DPA, which makes self-hosting a sales advantage rather than a cost.

Common questions

Does FERPA allow us to use AI on student records?

It allows disclosure to a vendor under the school official exception, which requires direct control over the vendor's use of the data. Self-hosting keeps the record inside the institution, so the exception is not needed and the analysis is much shorter.

What about students under 13?

COPPA attaches verifiable parental consent to collecting their personal information, and most general-purpose AI terms of service explicitly disclaim suitability for that age group. This is the constraint that usually decides the architecture in K-12.

Will this be used to grade students?

We do not build systems that assign grades. Assembling evidence and drafting feedback for an instructor to review is a different thing from making the assessment, and the distinction matters both educationally and for IDEA and 504 obligations.

Staff are already using ChatGPT on student work.

Almost universally true, and almost never documented. The first step is finding out what is actually in use, because governance written without that knowledge tends to prohibit the things people depend on and get routed around.

How we build it

Local & On-Prem LLM Deployment

The most powerful AI models, running entirely on your hardware.

AI Strategy & Roadmap

A clear path from where you are to where AI can take you.

Training & Change Management

Technology alone doesn't transform organizations, people do.

Internal Knowledge Base

Answers from your own documents, under your existing permissions.

AI Data Privacy

Keeping regulated data out of models you do not control.

Customer Service AI

Deflect the routine volume without trapping anyone in a loop.

Start with a conversation, not a proposal

Thirty minutes. We will tell you what we would change first, and whether you need us at all.

Book a call

The firm behind the firm