Recurvia develops intelligent systems and tooling that enable self‑optimising AI models, focusing on continuous optimization rather than static training pipelines. Its flagship product, Paramorph, uses adaptive control and multi‑agent learning to dynamically adjust hyperparameters and training behavior at runtime, improving convergence speed, scalability, and compute efficiency for large‑scale neural networks. By accelerating iteration velocity and reducing GPU and memory bottlenecks, Recurvia helps AI teams deploy more performant models faster.
Funding
$1M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
Training large AI models often relies on static hyperparameter settings and fixed training pipelines, leading to inefficient compute usage, long convergence times, and extensive manual hyperparameter sweeps. As model sizes grow, these inefficiencies increase pressure on GPU resources and infrastructure, slowing iteration cycles.
Solution
Recurvia offers a suite of intelligent systems that enable self‑optimising AI models. Its core product, Paramorph, applies adaptive control and multi‑agent learning to adjust hyperparameters and training behavior in real time, turning optimization into a continuous, data‑driven process. By dynamically tuning optimization strategies during training, Paramorph improves convergence speed, reduces the need for exhaustive hyperparameter searches, and enhances overall compute efficiency for large‑scale workloads. Complementing Paramorph, the Metrana platform provides AI‑native observability, large‑scale metric logging, and agentic experimentation to monitor and steer complex training jobs. Together, these tools allow AI teams to accelerate iteration velocity while lowering infrastructure costs.
Target Audience
Primary customers are machine‑learning researchers, AI engineers, and infrastructure teams building large‑scale neural networks who need faster convergence and more efficient compute utilization.
Features
- Real‑time hyperparameter adjustment and layer‑wise adaptive control during model training
- Multi‑agent coordination that selects optimization strategies based on ongoing performance signals
- Automatic reduction of hyperparameter sweep overhead through continuous learning
- Integrated observability platform (Metrana) with large‑scale metric collection and intelligent analysis
- Agentic experimentation framework for automated testing of training configurations
- Compatibility with common deep‑learning frameworks and GPU clusters for seamless integration