AI safety + compliance for education
The eval + guardrail + red-team + audit platform for K-12, higher-ed, and EdTech vendors. Block FERPA-protected student-record leakage at the gateway, ship age-appropriate content scorers for under-13 surfaces, satisfy district + IRB audit obligations.
What ships today
Honest posture, not roadmap promises
Every checked item is in production today. In-progress items are flagged explicitly — no overclaiming, no vapor.
Built for buyer reality
EdTech AI use cases we ship for
AI tutoring chatbot
Student asks the AI for help on a homework problem. Must not give the answer outright (cheating), must redirect appropriately when off-topic, must catch crisis signals (self-harm / abuse disclosures) and escalate.
EvalGuard features
- Topic firewall: blocks queries asking for direct homework answers (Socratic-mode default)
- Safety scorer (judge-safety): flags self-harm / abuse-disclosure signals and routes to escalation
- Age-appropriate content filter: per-grade-band thresholds for toxicity / violence / adult content
- Audit log: privacy-preserving conversation log for IRB + parental review
Grading + feedback automation
AI drafts feedback on student essays. Must not leak prior students' work, must give substantive feedback grounded in the rubric, must satisfy FERPA on student-record handling.
EvalGuard features
- PII firewall: student name + ID + grade-history auto-redacted from logs
- Faithfulness scorer: every feedback line grounded in a specific rubric criterion
- Output toxicity scorer: catches dismissive or judgmental phrasing in essay feedback
- Cross-tenant RLS: per-classroom isolation enforced at every DB query
Student-facing search + research
AI helps students find primary sources for a research paper. Must cite real sources (not hallucinate citations), must respect age-appropriate content filters, must catch academic-integrity red flags.
EvalGuard features
- Citation-format scorer: flags citations that aren't well-formed or grounded in real sources
- Hallucination guard: faithfulness check against the institutional library
- Semantic-similarity scorer: flags responses that closely echo source text (potential-plagiarism signal)
- Per-school BYOK keys: institutional encryption isolates research per university
District-IT chatbot for parents/staff
School-district chatbot answers parent + staff questions about enrollment, schedules, transportation. Cannot leak student records (FERPA), cannot give legal-style accommodation determinations.
EvalGuard features
- FERPA firewall: student record fields redacted before any LLM call
- Topic firewall: blocks queries asking for IEP / 504 / disability determinations
- Per-district rate limits prevent enumeration attacks against the SIS API
- Audit log: every interaction tagged with district + program for OIG audit recoverability
Wire it in 60 seconds
Wrap your OpenAI client. Scope to a district + grade band.
Per-district project IDs isolate FERPA boundary + age-appropriate scorer thresholds. Safety-scorer violations (e.g. self-harm signals) fire an onOutputViolation callback that escalates to district counselor workflow.
import OpenAI from "openai";
import { wrapOpenAI } from "@evalguard/openai";
const openai = wrapOpenAI(new OpenAI(), {
apiKey: process.env.EVALGUARD_API_KEY!,
projectId: `tutor_${districtId}_${gradeBand}`,
metadata: { vertical: "edtech", districtId, gradeBand },
blockOnViolation: true, // student PII pre-egress block
evalOnResponse: { failOnScore: 0.9 }, // age-appropriate + safety gate
onOutputViolation: (s) => escalateToCounselor(s, districtId),
});
await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: tutoringPrompt }],
});wrapOpenAI for wrapAnthropic.Stack
Six surfaces, one platform
Eval, firewall, red-team, audit, BYOK, dashboard — every surface ships out of the box. No bolt-on vendors, no procurement cycle per capability.
Ship classroom AI that parents trust + districts approve.
Free trial includes FERPA firewall, age-appropriate scorers, safety (self-harm) detection, and the full audit log. District-level deployment on request.
Apache-2.0 source · SOC 2 evidence engine live · full trust center