Summary
Dani Balcells is an AI safety research engineer with a decade of experience building and deploying ML systems, currently designing dangerous-capability and preparedness evaluations at OpenAI. With a background bridging mechanistic interpretability and product engineering, he has shipped LLM-powered features, fine-tuned proprietary models, and promoted eval-driven development as Plastic Labs’ founding ML research engineer. Prior to that he spent over six years at BMAT, where he progressed from developer to launching and running the US business unit and owning high-throughput music rights products processing hundreds of millions to billions of transactions. He combines deep hands-on skills—implementing transformers, GANs, RLHF and optimizing large-scale PyTorch workflows—with strong product instincts and stakeholder experience across C-suite and industry standards groups. Dani’s current mechanistic interpretability work probes how models form internal concept geometries at production scale, reflecting a rare mix of safety-focused research and pragmatic, production-hardened ML engineering.
10 years of coding experience
8 years of employment as a software developer
Communications Engineering - Signal Processing Computer Science, Communications Engineering - Signal Processing Computer Science at UPC - ETSETB TelecomBCN
Master of Science (MSc) Sound and Music Computing, Master of Science (MSc) Sound and Music Computing at Universitat Pompeu Fabra
Science and Technology Baccalaureat High School Degree, Science and Technology Baccalaureat High School Degree at IES Canarias Cabrera Pinto
English, Catalan, Spanish, French