Nilou Salehi
AI researcher and UC Berkeley professor. Her research centers on human-centered AI, human-computer interaction, and reliable AI systems in high-stakes settings.
We started Across AI to solve the hardest problem in enterprise AI: long-horizon reasoning for agents.
Across comes from years of public research on memory, reasoning, coordination, and AI systems for high-stakes work. The company brings that research together with enterprise platform experience.
AI researcher and UC Berkeley professor. Her research centers on human-centered AI, human-computer interaction, and reliable AI systems in high-stakes settings.
Dual PhD, ML algorithm designer, and USC professor. His work brings a systems and coordination lens to the architecture of enterprise agents.
Niloufar Salehi on reliability, context, and why agentic systems require new data structures.
Selected papers across HCI, social computing, knowledge work, and reliable AI systems.
Course material and research framing around design, development, and evaluation of AI systems.
Across combines research, product, engineering, design, finance, and enterprise operating depth, guided by advisors and backers who have built category-defining systems.
Leadership
Founding members & engineers
Founding UX

Founding Engineer


Founding Engineer


Founding Product Design

Founding Engineer
Founding Engineer
Founding Engineer
Head of Engineering


Founding Engineer
Finance / HR
Advisors & investors
Vas Bhandarkar
Founder/CEO ScoreData
Co-founder Selectica ($5B IPO).
Jeff Dean
Chief Scientist, Google DeepMind
Co-designer of MapReduce and TensorFlow.
Frederic Kerrest
Co-Founder & former COO, Okta
Grew Okta to a $40B+ public company.
Steven Mih
Former CEO Aviatrix, co-founder Ahana Cloud.
"Tackling one of the hardest problems in enterprise AI."
"Has the potential to transform how enterprises deploy AI."
Across convenes two audiences: enterprise leaders moving from pilots to production, and researchers working on AI systems that learn from interaction.


Bring Across into one workflow where memory matters, review matters, and the next action cannot be guessed from a single prompt: payables, a sprint team, a book of accounts.