Changelog

Unified Multi-Tenant AI Infrastructure

Migrated from per-tenant Kubernetes pod architecture to shared inference with dedicated Weaviate vector stores. Each organization now gets logical isolation via native Weaviate multi-tenancy with automatic pod isolation upgrade for enterprise workloads.

Technical Stack

• •Model: Chiro (state-of-the-art language model)
• •Inference: Shared inference service (OpenAI-compatible API)
• •Vector Store: Weaviate with multi-tenant collections
• •Fine-tuning: LoRA adapters with hot-loading
• •Development: Local development inference with automatic API detection

Tier-Based Quotas

• •Spark: 2048 tokens/req, 10k tokens/hr, 5 concurrent requests
• •Seed: 4096 tokens/req, 100k tokens/hr, 20 concurrent requests
• •Growth: 8192 tokens/req, unlimited tokens/hr, 100 concurrent requests
• •Dedicated: 16384 tokens/req, unlimited, unlimited (auto pod isolation at $10k+/month)

RAG Pipeline

512-token document chunks with 50-token overlap, hybrid BM25+vector semantic search, automatic context injection before inference.

Pod Isolation Triggers

Elite tier ($10k+/mo), 80%+ quota utilization for 1hr, compliance requirements (HIPAA/SOC2), or P99 latency >500ms for 30min.