Intelligence, Compounding

Recurvia is a deep reinforcement learning lab building AI that improves AI. We are the observation, reasoning, and optimisation loop of self-improving intelligence.

About Us

Built by researchers and engineers from

Educated at

Our DNA

Our work is driven by a single research question: how can AI systems become better at improving themselves, and by extension at solving many of the world's hardest problems? The technologies we build resolve to one unified answer: a closed loop in which AI observes how models learn, reasons about what to run next, and acts on training in real time.

Isometric illustration of a microprocessor unit

Building AI that can keep up

Training and deploying advanced AI systems increasingly depends on iteration velocity and compute efficiency — at every level of the stack.

3D Illustration of a convolution layer diagram

Training will learn to run itself.

Demand for compute continues to outrun supply across GPUs, memory, and large-scale infrastructure. Every wasted cycle now has a price — and no team can watch, diagnose, and retune its runs fast enough to reclaim it by hand. The answer is AI systems that analyse, optimise, and adapt themselves — continuously, observably, and within limits their operators set.

One system. First layer live. The loop ahead.

Everything we build forms a single loop. Metrana is the observation layer, already live: high-fidelity telemetry on how models learn, designed to be read by machines as well as humans. Paramorph is the optimisation layer: dynamic control of training at runtime. Connecting them is a growing reasoning capability — automated analysis and experimentation that decides what to run next. One system, designed around continuous optimisation rather than static workflows.

Metrana

The observation layer for self-improving AI.

Built first for our own research, Metrana is an observability and experimentation platform designed for a world in which the primary consumer of training telemetry is increasingly an AI system itself.

Metrana combines metric logging at frontier scale with intelligent analysis and agentic experimentation: the foundations of the loop's reasoning layer, available today to the teams training the most demanding models.

LEARN MORE

Metric logging at massive scale

RL-specific observability interfaces

AI-assisted analysis

Automated experimentation and iteration

PARAMORPH

A foundational AI for autonomous neural network optimisation

Paramorph is the loop's optimisation layer. It adapts learning strategies live during training — continuously adjusting hyperparameters and training behaviour to improve convergence, scalability, and compute efficiency. Itself a foundation model trained with multi-agent reinforcement learning, Paramorph treats optimisation as an intelligent, continuously evolving process rather than a static configuration problem.

LEARN MORE