Ratio is a cloud-native AI composition platform built on AWS. It provides a versioned filesystem, event-driven tool execution, and a dependency-aware orchestration engine for building autonomous AI systems with built-in safety controls.
Why I built it
I wanted a system where AI processes could manage and evolve functionality independently, but within an architecture that kept them controlled. The idea was that AI should be able to compose new capabilities by combining existing tools in new ways, building on itself, but only within defined boundaries.
That meant the platform needed real infrastructure: a versioned filesystem that tracks every data change, an auth system that controls what has access to what, event subscriptions that trigger automation reactively, and an orchestration layer that resolves dependencies between tools automatically. If an AI process is going to operate autonomously, the system around it has to be engineered for safety.
How it works
Tools are the core unit. A simple tool is a single function. A composite tool chains multiple tools together with automatic dependency resolution through a REF system, where outputs from one step feed into the next. Composite tools support parallel execution over collections, conditional logic, and unlimited nesting depth, so you can compose complex workflows from smaller pieces.
The filesystem is versioned. Every file has lineage tracking and Unix-like permissions. Event subscriptions let tools fire automatically when something changes in the system. The auth layer uses RSA key-based entity and group management to control access to resources.
Everything runs on Lambda, S3, DynamoDB, and SQS. Serverless and pay-per-use, so it scales to zero between runs and costs almost nothing to operate.
The platform worked well. I've mostly moved on from it as I haven't had a major use case recently, but the thinking behind it, controlled autonomy, composable tooling, safety by architecture, continues to inform how I approach AI systems.