Infrastructure Foundation

In a single sprint we provisioned the complete Intellixer production stack: Google Cloud infrastructure via Terraform (GCE VM, Cloud SQL Postgres, Secret Manager, GCS backups, KMS encryption), a LiteLLM API gateway with Postgres-backed usage tracking, Presidio-based PII anonymisation pipeline, and the on-prem Mac Mini M4 inference node running MLX via a custom FastAPI shim behind Caddy TLS.

Architecture

Client → api.intellixer.farm (GCE, LiteLLM) → dc1.webhop.me (Caddy/Mac M4) → mlx_lm

• 65 infrastructure files committed in a single day
• Terraform modules: GCE, Cloud SQL, GCS, KMS, IAM, Secret Manager
• LiteLLM proxy with custom callbacks for audit + privacy
• Caddy reverse proxy with token-authenticated on-prem shim
• Presidio NLP pipeline for GDPR PII scrubbing before inference

Hardware

Mac Mini M4 (16 GB Unified Memory) runs quantised 3B-parameter models at 30+ tokens/second on the Apple Neural Engine — at a fraction of GPU cloud cost.