Your campus runs on Google Cloud. Your email is Gmail, your collaboration suite is Google Workspace, and your data warehouse is BigQuery. When it comes time to add generative AI for student support, the last thing you need is a vendor that drags you into a different cloud.

ibl.ai deploys natively on Google Cloud Platform, using Vertex AI for model inference, GKE for container orchestration, and the full GCP security and observability stack. Your data stays in your project. Your IAM policies stay intact. Your billing stays transparent.

---

Vertex AI: The Model Backbone

Vertex AI is Google Cloud's unified machine learning platform, and it has become one of the most comprehensive model serving environments in the industry. Through the Vertex AI Model Garden, institutions get single-API access to:

• Gemini 2.0 Flash — Google's fastest multimodal model, ideal for real-time tutoring with sub-second latency.

• Gemini 1.5 Pro — With up to 2 million tokens of context, perfect for analyzing entire course syllabi, research papers, or lengthy student portfolios in a single pass.

• Gemma — Google's open-weight model family, available for fine-tuning on institutional data directly in Vertex AI.

• Llama 3, Mistral, and others — Third-party models hosted in Model Garden, giving institutions choice without leaving the GCP ecosystem.

ibl.ai's orchestration engine treats Vertex AI as a first-class model provider. Each AI agent can be configured to route different tasks to different models:

Task	Recommended Model	Why
Live tutoring chat	Gemini 2.0 Flash	Sub-second latency, multimodal
Deep analysis / long docs	Gemini 1.5 Pro	2M token context window
Cost-sensitive batch jobs	Gemma or Llama 3	Lower per-token cost
Domain-specific tutoring	Fine-tuned Gemma	Institutional knowledge baked in

Switching models is a configuration change—not a code change. When Google releases the next Gemini version, ibl.ai picks it up immediately.

---

Vertex AI Agent Builder and RAG Engine

Raw model intelligence is not enough for education. Answers must be grounded in your curriculum—your syllabi, your textbooks, your institutional policies.

Vertex AI RAG Engine (part of Agent Builder) provides managed retrieval-augmented generation:

1. Ingest course materials from Cloud Storage or Google Drive into a managed vector store. 2. Retrieve relevant passages at query time—automatically, with no custom retrieval code. 3. Ground model responses in actual course content, reducing hallucination and increasing pedagogical accuracy.

ibl.ai's agents chain RAG retrieval with multi-step reasoning. A student asking about enzyme kinetics gets an answer grounded in their actual biochemistry textbook—not a generic internet summary.

---

Infrastructure: GKE, Cloud SQL, and the Familiar Stack

ibl.ai's microservices run on Google Kubernetes Engine (GKE) Autopilot, Google's fully managed Kubernetes offering. Autopilot eliminates node management—you pay only for the pods your workloads consume.

The deployment includes:

• GKE Autopilot — API, orchestration engine, background workers, all auto-scaled.

• Cloud SQL (PostgreSQL) — Relational store for user data, session logs, and course mappings. Multi-tenant schemas or per-database isolation.

• Cloud Storage — Object store for lecture materials, embeddings, model artifacts, and backups. Partitioned by tenant prefix.

• Memorystore (Redis) — Session caching and real-time feature serving.

• Cloud Monitoring and Cloud Logging — Unified dashboards for infrastructure health, model latency, and cost tracking.

Everything runs inside your VPC with private Google Access. Vertex AI connections stay on Google's backbone—no public internet traversal.

---

Security: VPC Service Controls and Beyond

Google Cloud's VPC Service Controls create a security perimeter around your GCP resources. For ibl.ai, this means:

• No data exfiltration. Even if a service account is compromised, VPC Service Controls prevent data from leaving the perimeter.

• IAM + Organization Policies. Least-privilege roles for every ibl.ai microservice. Organization-level constraints prevent accidental misconfigurations.

• Encryption everywhere. Data encrypted at rest (Google-managed or CMEK) and in transit (TLS 1.3). Customer-managed encryption keys (CMEK) available for institutions that require key custody.

• Audit logging. Cloud Audit Logs capture every API call for compliance review. Export to BigQuery for long-term analysis.

FERPA Alignment

Student data stays inside the institution's GCP project. Vertex AI processes prompts ephemerally—Google does not use customer data to train models. Combined with VPC Service Controls and IAM, institutions can demonstrate a defensible FERPA compliance posture.

---

BigQuery Integration: Analytics That Close the Loop

Many GCP-native institutions already run their data warehouse on BigQuery. ibl.ai writes learning analytics—session summaries, topic mastery signals, engagement scores—directly to BigQuery tables.

This means your institutional research team can:

• Join AI tutoring data with enrollment, retention, and financial aid data in a single SQL query.

• Build Looker dashboards that show AI impact alongside traditional student success metrics.

• Run predictive models in BigQuery ML to identify at-risk students, using tutoring engagement as a feature.

The data flows in both directions. BigQuery data (student demographics, prior academic history) can inform ibl.ai's agents through Vertex AI RAG Engine—closing the loop between analytics and action.

---

Cost Governance

Because ibl.ai runs in your GCP project, every dollar is visible in Cloud Billing:

• Vertex AI charges per token (input and output), broken down by model.

• GKE Autopilot charges per pod resource-second—no idle node costs.

• Cloud SQL and Cloud Storage follow standard GCP pricing with committed-use discounts.

ibl.ai's admin console adds educational context: cost per student, cost per tutoring session, cost per department. Budget alerts in Cloud Billing catch anomalies before they become surprises.

---

Getting Started

1. Create a dedicated GCP project (or use an existing one) for ibl.ai workloads. 2. Enable Vertex AI APIs and request access to your preferred models in Model Garden. 3. Deploy via Terraform. ibl.ai provides IaC modules for GKE, Cloud SQL, Cloud Storage, and IAM—your security team reviews before applying. 4. Connect your LMS. LTI 1.3 integration with Canvas, Blackboard, Moodle, and Open edX. 5. Ingest course materials. Upload to Cloud Storage; ibl.ai indexes them via Vertex AI RAG Engine. 6. Pilot, measure, scale. Start with one department. Measure engagement, outcomes, and cost. Expand.

---

The Bottom Line

If Google Cloud is your institution's home, ibl.ai moves in—not next door. Your VPC, your IAM, your billing, your BigQuery warehouse—all intact and enhanced with AI-powered tutoring, advising, and content creation that runs on the same infrastructure you already trust.

Ready to deploy ibl.ai on Google Cloud? [Contact us](https://ibl.ai/contact) for a technical architecture session.