Sepehr Akhavan

Software Engineer at Meta

My work and life are guided by the same principle: experience becomes meaningful when we learn from it. In AI, I study how AI systems become capable and reliable after pretraining, through agentic post-training, context and memory, harness design, interaction feedback, evaluation, and continual self-evolution. In life, I practice being present and reflecting on experience, growing with intention and gratitude for the priceless gift of being alive.

I am a Software Engineer at Meta, where I build self-improving AI systems. My broader contributions at Meta span agent evaluation and observability, continual learning, AI-assisted software engineering, and adaptive model optimization. Previously, I worked on Bean Machine, an open-source universal probabilistic programming language for fast and accurate Bayesian analysis. Before joining Meta, I spent nine years at two startups: Cylance (acquired; 2016–2019) and MIND Research Institute (2010–2016). I earned my B.S. from Sharif University of Technology and my M.S. and Ph.D. from UC Irvine, where my doctoral research focused on Bayesian nonparametric modeling.