Last Updated on June, 29, 2026
Purpose
Disclose to customers and end users how CampusAI uses AI, what data flows to the language model, and the limits of model output.
AI components in CampusAI (verified)
● Chat agents (10 sub-agents + root) — answer questions, provide recommendations, surface campus information
● Resume recommendations (ATS-first suggestions only; CampusAI does not generate a resume document)
● Personalized feed (My Feed)
● Weekly digest (batch-fired)
● Crisis-detection / wellness routing with hardcoded crisis resources fallback (988 / Crisis Text Line / university counseling)
Models Used
● All chat agents: gemini-3.5-flash (Google Cloud Gemini API)
● Embeddings: gemini-embedding-001, 768-dim
What is Sent to the Model
● Conversation content the user types
● Profile + preferences retrieved from the tenant-isolated user store (program / major / year / interests)
● Catalog references (course / event / job text) retrieved from uip-* serving tables and the pgvector catalog
● Adaptive memory context from ACE playbook and LCM episodic memory, both per-tenant per-user
What is Not Sent to the Model
● Other tenants' data (enforced by the 4-layer tenant isolation; see Data Classification Map)
● Raw payment, government identifier, or third-party credential material
● PII that the user did not intentionally introduce (the PIIRedactionPlugin runs first in the agent pipeline)
What the Model is Not Used For
● Foundation-model training - as per Google Cloud Vertex AI Gemini DPA
● Cross-tenant inference or recommendations
● Decisions about employment, admission, financial aid, or other consequential outcomes — recommendations are advisory only
Limitations User Should Know
● AI output may be inaccurate or incomplete
● Recommendations are not professional advice
● Wellness information is not medical or therapeutic advice; crisis resources are surfaced but are not a substitute for professional care
Where This Disclosure Appears
● Terms of Service (campusai.app/terms) — §5
● Privacy Policy (campusai.app/privacy) — §4
● In-product onboarding (first-run disclosure)
● Customer contract addendum — incorporated into the customer agreement as the AI Use addendum.
Approved by: Aiden Polinsky, Founder & CEO
UniversityAI, Inc.
Signature: __________________________ Date: 2026-06-29