RL Environments for
Code Generation.

metaphi is an applied research lab building infrastructure to scale out RL environments where agents can learn from dynamic user persona simulation for code generation tasks.

we are a team of ML engineers, founders and RL researchers who obsess about simulations and variant design for expanding humanity's intelligence.

today agents generate predictions that are largely one size fits all. this is due to the missing agent training layer where agents can dynamically incorporate explicit and implicit user preferences for personalized predictions and task execution.

based in San Francisco, CA. just like our agents, we love our time in the training gyms too.

Contact Us

Research Focus

Dynamic User Simulation.

building realistic user personas that interact with generated interfaces in real-time, creating diverse behavioral patterns and edge cases that expose the limitations of static code generation approaches.

Adaptive Code Generation.

developing agents that learn from user interaction patterns to generate personalized UI components, adapting both visual design and functional behavior based on observed user preferences and workflows.

Reinforcement Learning Infrastructure.

creating scalable environments where code generation agents receive continuous feedback signals from simulated users, enabling iterative improvement through reward modeling and policy optimization.