Niklas Nielsen is co-founder and chief technology officer of Log10, creator of Everest, an AI platform for clinical, regulatory, and other high-stakes life-sciences documentation. He develops systems that make generative AI dependable through application-specific testing, expert feedback, and auditable professional workflows.
Nielsen previously worked on virtual machines and compilers at Adobe and distributed systems at Mesosphere and Intel. An Apache Mesos committer, he authored the project’s 2015 release announcement, detailing improvements to cluster maintenance, resource reservations, and container networking. He subsequently co-founded CustomerDB, an AI-driven product-management startup, and became head of product at MosaicML, where he highlighted Composer’s model-training performance tools.
Nielsen founded Log10 with Arjun Bansal in 2023. The company raised $7.2 million in seed funding in 2024 and initially built tools for monitoring model outputs, collecting feedback, debugging failures, and improving training data. Its Everest platform applies those capabilities to regulated life-sciences documentation.
- Application-specific model evaluation: Nielsen developed llmeval, a local Python framework combining Hydra-based configuration, customizable checks, and cross-model reporting. His AI Engineer demonstration showed how extra explanations, unexpected whitespace, or unparseable Python can make otherwise plausible outputs unusable.
- Reliability across repeated trials: Because model responses vary, llmeval evaluates multiple samples against configurable passing thresholds, making prompt changes and model-provider comparisons measurable.
- Human-guided model evaluation: Nielsen combines rubric-based automated grading with existing reviewer feedback. He accounts for weaknesses in model judging—including preference for a model’s own outputs and unreliable numerical scores—while preserving domain experts’ judgments in sensitive documentation workflows.
