How the platform uses AI — and the guardrails around it.
Assistants answer only from approved material and present citations. When confidence is low, the platform escalates to a person instead of guessing, and every AI action is written to the audit log.
Demonstration software. Pre-build, pre-clearance. No FedRAMP / StateRAMP / SOC 2 / Protected B / accessibility conformance is claimed, and the platform is not in government production use. All demo data is fictional. CivicMatrix AI and Public Agency OS are working product names of LTI Group LLC.
Grounded and cited
Assistants answer only from approved source material and present citations. They are instructed to say when they do not know and to point to a person, rather than to speculate.
Human in the loop
When confidence is low, when retrieval is weak, or when a question falls outside approved material, the platform escalates to a person and records the escalation. Consequential workflow steps are gated on human approval.
Safety screening
Emergency and prohibited-topic prompts are screened before any model call; prompt-injection attempts in user input and retrieved content are detected and treated as data, never instructions.
Data & governance
By default the platform does not train models on customer data. A central gateway enforces an allow-list of models, timeouts, a circuit-breaker, and per-agency spend caps, and it fails closed to human escalation on any model problem. Every AI action is written to the append-only audit log.
Limitations
AI output can be wrong or incomplete; it supports, and does not replace, professional judgment. Answer quality depends on the approved sources provided and the configured model. Nothing here claims certification, government approval, or production readiness.
Contact
Questions: govtech@ltigroupllc.com