Enterprise Security & Architecture Overview

Workloads execute within containerized serverless microservices across Google Cloud Run and AWS Lambda (US-East) inside private Virtual Private Clouds (VPCs) with zero public interface exposure.

Managed Cloud SQL utilizing pgvector for vector similarity search, logically partitioned at the database schema level to guarantee multi-tenant data isolation.

LiteLLM Proxy manages rate limits, cost tracking, and secure egress to foundational models via enterprise-tier Google Vertex AI endpoints.

100% of cloud resources are provisioned, version-controlled, and audited using modular Terraform scripts.

Customer inputs, prompts, vector embeddings, and outputs are contractually and technically excluded from training, fine-tuning, or aligning public or multi-tenant models.

Model inference calls operate over enterprise Vertex AI APIs under strict zero-prompt-retention terms.

Upon subscription termination or written request, all tenant vector stores, database schemas, and backups are permanently purged within 30 days.

Full TLS 1.3 encryption in transit across all microservices and API gateways; AES-256 encryption at rest managed through Google Cloud Key Management Service (KMS).

Secrets, database connection strings, and model credentials are store-encrypted and injected dynamically at runtime via Google Cloud Secret Manager.

Principle of Least Privilege (PoLP) enforced through GCP Identity & Access Management (IAM) and Role-Based Access Control (RBAC).

All administrative actions, code deployments, and API system events are recorded via Cloud Audit Logs for continuous security monitoring.

Built on AWS / GCP hybrid infrastructure adhering to SOC 1/2/3, ISO/IEC 27001, and PCI-DSS security frameworks. GCP is HIPAA BAA-eligible; MarketMind AI is not currently HIPAA-compliant — PHI processing requires an executed BAA (see AUP).