About Sleepers
Research.
An independent laboratory investigating structural alignment, agentic safety protocols, and robust middleware systems.
First Principles
Sleepers Research exists as a technical sandbox for understanding machine learning from first principles rather than building surface-level product wrappers. The focus has always been on the core mechanics of how transformer models actually work, rather than just how to prompt them.
Foundational architectures like GPT-2 have been rebuilt from scratch here, focusing on attention distribution, weight matrices, and the internals most people skip past. That groundwork shaped how the lab approaches fine-tuning, alignment, and system safety: by understanding what's actually happening inside a model, not just what comes out of it.
Today, the lab designs secure middleware pipelines for autonomous systems. This infrastructure is built to sit between capable AI agents and the environments they operate in. Sleepers Research is a builder-led effort, not an institution: small, deliberate, and focused on problems that matter as agentic systems become more capable and more autonomous.
Lab Philosophy
Depth Over Surface
Every system here is built to understand what's happening underneath, including the model, the data, and the runtime, rather than just sitting on top of it. Security and grounding operate inline, at the network and runtime layer.
First Principles
Every system we deploy is born from fundamental research. We inspect parameter drifts, token distributions, and system traces to construct robust defense vectors.
Latency Limits
Security shouldn't choke execution speed. Our systems are designed with high-throughput proxies (Rust-based) to enforce safety policies in sub-millisecond ranges.