Agent evaluation for education teams shipping AI
An agent tutoring a student or touching a student record is handling data with its own legal protections, and most schools have no record of what it actually accessed.
Every run scored for quality, performance and risk, and checked against the activity schema you define.
Every education agent run gets watched, evaluated for quality and risk, and checked against the rules you set. One record answers what any agent did, with evidence ready for review.
What's at stake in education
For the instructors, admissions officers, and edtech teams behind a tutoring or admissions agent, the stakes are student records the agent can reach and answers a student will act on as if they were correct.
Agents reach records with their own legal protections, and minors are often involved.
A wrong answer sends a student down the wrong path, and it reads as correct.
Few schools can show, for a given answer, which course material it actually came from.
One source of truth for your education agents.
Instrument the agent frameworks you build on, and ingest the systems those agents touch as custom spans. Every run, score and signal lands in one place.
Claude Agent SDK
LangChainSDKCore SDK · TS + PythonOTLPOpenTelemetry · closed toolsObserve. Evaluate. Act.
A record of what the agent did, and a check before it goes further. Each step below closes one of the problems above.
- Observe. Every record and source read recorded as its own span
- Evaluate. Each answer checked against the material actually retrieved
- Act. Written to an immutable audit trail, queryable per interaction
- FERPA
- COPPA
- GDPR (international students)
- State student data privacy laws
Built for everyone who owns education AI.
The same record, read the way each team needs it.
Instrument the education agents you already run — no gateway in the request path, no re-architecture.
For developers →Ship education features without regressions — every run scored before a customer ever sees it.
For product teams →One portfolio view of every education agent — its owner, cost and risk in a single place.
For heads of AI →Audit-ready evidence for every decision a education agent makes, ready when a regulator asks.
Security & governance →Two agents education teams are already shipping.
Hypothetical, but grounded in how education teams deploy agents today.
Tutoring assistant
Low risk · 89/100A Claude Agent SDK agent answers a student's homework questions by pulling from the course's approved textbook and lecture material, escalating to a human instructor when a question falls outside that material.
- Each answer the agent gives recorded as its own span
- Validated against an activity schema that checks the response is grounded in the course material it retrieved, rather than an unrelated or fabricated source
- Scored for quality and risk on every run
An instructor sees which course material backed a specific answer, and every time the agent escalated instead of guessing.
Admissions review assistant
Medium risk · 84/100A LangChain agent reviews applicant files in the student information system, summarizes each application against admissions criteria, and drafts a recommendation for an admissions officer to review.
- Every file the agent accesses and every recommendation it drafts captured as a span
- Validated against an activity schema that confirms the agent only read fields it was scoped to see for that stage of review
- A configurable risk profile flags a run that accesses a restricted field, such as financial aid data during an early review stage
An admissions office can confirm the agent read only the fields it was scoped to see at each stage, and stop a run that did not.
Runs stay isolated — terminate one without touching the rest of your fleet via the kill switch → · Prefactor vs. observability tools →
The problems this solves
Agents fail quietly and the first signal is a complaint, not an alert.
How it gets caught →Stuck at POCThe pilot worked; sign-off takes months because risk has no evidence.
How it gets caught →No kill switchWhen an agent misbehaves, nothing can stop it short of stopping everything.
How it gets caught →See agent evaluation on your own education agents
Book a demo and we'll walk through span-level scoring and audit evidence on a fleet like yours.
Unified performance platform for agents, authentication, and risk management