Listing + screening AI that survives a fair-housing audit
Listing copy that hints at a preferred tenant, a chatbot that steers by neighborhood demographics, a screening model with disparate impact — each is a Fair Housing Act case. EvalGuard gates every output on fair-housing scorers and keeps the decision trail HUD-style complaints demand.
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
Real estate AI use cases we ship for
Listing-copy generation at portfolio scale
AI writes hundreds of listing descriptions a day. Phrases like 'perfect for young professionals' or 'family-friendly block' read as preference signals under the FHA — every description needs a fair-housing gate, not a spot check.
EvalGuard features
- Steering-language guardrails block preference and demographic signals
- Bias scorers screen copy against protected-class proxy phrases
- Output guardrails strip preference language so descriptions stay amenity-first
- Audit log preserves every generated description for complaint defense
Tenant + buyer screening assistance
AI summarizes applications and flags risk factors. Any systematic disparity across protected classes is an ECOA/FHA exposure — screening needs disparity testing before deploy and a per-decision trail after.
EvalGuard features
- Bias scorers gate screening outputs; red-team packs run counterfactual disparity probes pre-deploy
- Applicant-PII firewall redacts identifiers before third-party model calls
- Tamper-evident audit log ties each recommendation to model + prompt version
- Evidence-bundle export packages the testing file for your counsel
Property-inquiry chatbot
Prospects ask about neighborhoods, schools, and availability. Answering 'is this a safe area?' with demographic composition is textbook steering — the bot must redirect to objective data sources every time.
EvalGuard features
- Steering guardrails block demographic-composition answers so the bot falls back to objective data sources
- Prompt-injection defense: 300+ attack plugins cover baiting the bot into steering language
- Faithfulness scorer grounds availability answers in live listing data
- Per-route rate limits contain scraping of portfolio availability
Lease + document Q&A for residents
Residents query an assistant about lease terms, fees, and notices. A hallucinated fee or invented clause becomes a landlord-tenant dispute — answers must quote the actual executed lease.
EvalGuard features
- Citation scorer requires lease-document grounding for every answer
- Hallucination gate blocks answers that drift from the executed terms
- Per-project BYOK isolation separates each property-management client
- Gateway cost ledger attributes spend per property, with per-tenant daily budget caps
Wire it in 60 seconds
Wrap your OpenAI client. Every listing passes fair housing.
Steering rules + bias thresholds + decision-trail 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: "listing-copywriter",
metadata: { vertical: "real-estate", fha: true },
blockOnViolation: true, // refuse steering language
evalOnResponse: { failOnScore: 0.7 }, // fair-housing gate
onViolation: (r) => alertComplianceDesk(r.violations),
});
try {
await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: listingBrief }],
});
} catch (err) {
if (err instanceof EvalGuardViolationError) {
// Block recorded in the decision 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 housing AI you can defend?
Free trial includes the fair-housing scorers, steering guardrails, and the decision audit trail. Bring your listing templates — we'll show you what a HUD complaint would find.
Apache-2.0 source · SOC 2 evidence engine live · full trust center