AI safety + compliance for life sciences
The eval + guardrail + red-team + audit platform for AI touching regulated pharma workflows. Catch hallucinated dosages and off-label claims before they ship, keep trial data out of third-party LLMs, and hand your quality team a tamper-evident audit trail built for GxP-style review.
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
Pharma & life sciences AI use cases we ship for
Medical-information response drafting
AI drafts responses to healthcare-professional inquiries from the approved label and med-info library. A hallucinated dosage, contraindication, or off-label claim is a reportable event — every response must stay inside the approved source set.
EvalGuard features
- Faithfulness scorer: every claim checked against the approved label + source library
- Citation-format scorer rejects responses that cite outside the approved corpus
- Output guardrail block mode: un-sourced dosage or efficacy claims never reach the requester
- Tamper-evident audit log: every draft + reviewer decision preserved for inspection
Pharmacovigilance intake triage
AI screens inbound emails, call transcripts, and social mentions for potential adverse events. Missing a signal is a compliance failure; patient identifiers must be redacted before any third-party model sees the text.
EvalGuard features
- PII firewall redacts patient name / DOB / contact details before the LLM call
- Toxicity + safety scorers flag candidate adverse-event language for human triage
- Answer-relevance scorer gates the triage summary against the source narrative
- Gateway cost ledger attributes every intake scan to the PV cost center
Clinical-trial protocol Q&A for site staff
Site coordinators query an assistant grounded in the protocol + investigator brochure. Wrong visit windows or dosing schedules create protocol deviations — answers must be faithful, versioned, and auditable per site.
EvalGuard features
- Hallucination scorer: answers checked against the current protocol version
- Prompt-injection defense: 300+ attack plugins cover extraction attempts against trial data
- Per-project BYOK isolation keeps each sponsor's trial corpus separate
- Audit log records protocol version + answer per site for deviation review
Regulatory-submission drafting support
Writers use AI to summarize study data into submission modules. Numbers must match the source tables exactly, and the quality team needs evidence of every AI touch on the document trail.
EvalGuard features
- Faithfulness scorer checks summary claims against the source tables
- Multi-model routing with quality-cost strategy — accuracy threshold enforced per section
- Evidence-bundle export: AI-assist trail packaged for quality + regulatory review
- Per-tenant daily budget caps stop runaway drafting jobs before they burn spend
Wire it in 60 seconds
Wrap your OpenAI client. Get label-grounded guardrails.
Faithfulness thresholds + redaction rules + audit retention are configured once in the EvalGuard control plane. Your code only wraps the client.
import OpenAI from "openai";
import { wrapOpenAI, EvalGuardViolationError } from "@evalguard/openai";
const openai = wrapOpenAI(new OpenAI(), {
apiKey: process.env.EVALGUARD_API_KEY!,
projectId: "med-info-responses",
metadata: { vertical: "pharma", gxp: true },
blockOnViolation: true, // refuse un-sourced claims
evalOnResponse: { failOnScore: 0.75 }, // label-faithfulness gate
onViolation: (r) => notifyQuality(r.violations),
});
try {
await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: medInfoPrompt }],
});
} catch (err) {
if (err instanceof EvalGuardViolationError) {
// Block recorded in the tamper-evident audit trail. Replay via audit ID.
}
}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.
Ready to ship regulated-pharma AI you can defend?
Free trial includes the data firewall, faithfulness scorers, and the tamper-evident audit trail. Self-hosted deploy available for validated environments.
Apache-2.0 source · SOC 2 evidence engine live · full trust center