#Routing

AI Gateway: From Backend Scheduling to Cost Governance

Using multi-backend model access as the setting, this article explains how an AI Gateway handles resource scheduling, capacity control, failure switching, distributed state, usage accounting, and enterprise cost governance.

Short

16 min read

Part of the column “Technical Systems” · Chapter 9

• #AI
• #Architecture
• #Systems Design

Where Exactly Is the Cache? Locality in a Multi-Layer Execution Service

Using a multi-layer execution service as an example, this article breaks down the multiple levels of locality across request identity, edge routing, execution resources, and backend caches, then presents reusable approaches to routing, invalidation, failover, and observability.

Short

13 min read

Part of the column “Technical Systems” · Chapter 7

• #Performance
• #Systems Design
• #Algorithms

Beyond Model Switching: Engineering Design and Routing Algorithms in DSH Quota Router

How candidate chains, failure classification, per-turn identity, and cost formulas make multi-source model routing controllable and auditable.

Quota Router: An Explainable Multi-Source Fallback Chain for DSH

A policy-only DeepSeek Harness plugin that builds ordered, explainable, auditable model candidate chains per task and advances along them within bounded failure rules.

Neo Matrix

An OpenAI-compatible relay platform with a revenue-sharing mechanism for individual suppliers: consumers call models through one unified entry point, suppliers host idle upstream keys, and the platform handles routing, metering, settlement, and withdrawals. This article introduces the project, its mock demo, business model, financial estimates, and the risks and sunk costs that must be accepted before launch.