Isidor

Isidor is an applied research and product company working on the foundation of reliable AI. Its systems verify AI outputs, benchmark performance and automate evaluation across billions of examples. The work spans AI reliability and evaluation, automated model benchmarking, reinforcement-learning data pipelines, data curation and post-processing, Gen AI output post-processing and workflow capture - producing cutting-edge performance on firm-specific workflows at scale.

Three lines of work define the company's output. Secure, vertically-integrated pipelines transform raw, unstructured proprietary data into reinforcement-learning-ready curricula for model training. Evaluation and verification systems measure and automate assessment of model behaviour at scale. Specialized benchmarks are tailored to individual industry verticals. The work is carried out in service of Enterprise AI and frontier AI labs.

Isidor is founded by engineers and researchers concerned with precision, transparency and scientific rigor, and it states an ambition to set a new standard for how AI reliability is measured and evaluated. For technical practitioners, the appeal lies in problems at the intersection of applied research and production systems: building evaluation infrastructure, data pipelines and benchmarking methods that hold up under scrutiny and under load.