

MatrAIx Neurips Party
MatrAIx (https://matraix.ai/) is an AI startup working at the model layer, founded by people from OpenAI, Google DeepMind, Microsoft, Harvard, and MIT. We build evaluation infrastructure that runs large populations of simulated users through personal AI agents and produces large-scale high-quality training RL data.
Our work centers on three areas:
RL Environments: environments and reward design for training agents
User Simulation and Synthetic Data: simulated users and synthetic training data
Personalization and Agent Evaluation: benchmarks for how agents adapt to different users, preferences, and contexts
We’re hiring. We’re looking for MatrAIx Fellows and full-time Members of Technical Staff in the three areas above. Research fellowships are open to PhD students and to researchers or engineers in industry. Roles are based in Mountain View, CA. Details are at https://matraix.ai/careers/.
Confirmed guests at the table:
▪ Senior Director at Apple
▪ Senior Director at OpenAI
▪ Senior Director at Microsoft AI
▪ Senior Staff Research Scientist at Google DeepMind
▪ Member of Technical Staff at xAI
Projects to check out:
PersonalAgentEval (https://personalagenteval.com/): a benchmark and leaderboard for commercial personal agents. It covers 152 tasks scored by capability and domain, and uses a simulated user who plays each person without seeing their accounts or the grader.
Personal Agents’ Last Exam (PALE) (https://personalagenteval.com/PALE): our open-source personal agent benchmark. Contributors build app or website environments or write personal scenarios, which then go through review. Contributors need to be affiliated with a North American institution.
Come talk to us about evals, training data, or joining the team!