About the job
Location: Remote or Palo Alto, CA
Duration: 12–16 weeks (flexible)
Compensation: Paid, competitive
Start: Rolling
About Palona
At Palona, we are dedicated to creating real-world AI systems that operate continuously in production environments. Our focus is on developing AI agents that can perceive, reason, remember, and act in physical spaces, starting with restaurants as our initial domain due to its complexity and high signal density.
We thrive on research that stands the test of reality, addressing challenges such as partial observability, delayed effects, noisy signals, non-stationarity, and long-term outcomes.
Research Scope
This internship is designed for PhD students eager to tackle applied research challenges linked to deployed systems. You will explore questions stemming from live AI agents functioning in the real world, where ideal assumptions may fail, and understanding system behavior over time is essential.
Required Research Background (PhD Level)
We seek candidates with extensive research experience in at least one primary area, alongside a working knowledge of related fields.
Primary Research Areas (at least one required)
1. Sequential Decision Making
- Reinforcement learning, planning, or control
- POMDPs or decision-making under partial observability
- Credit assignment with delayed and sparse rewards
- Long-horizon optimization
Relevant indicators:
- Publications in RL, planning, or control venues
- Experience in implementing and evaluating decision-making agents
2. World Modeling and State Representation
- Latent state models for dynamic environments
- Temporal abstraction and hierarchical representations
- Persistent memory or state tracking
- Modeling environments that evolve over time
- Research in state-space models, memory-augmented models, or temporal representations
3. Reasoning Under Uncertainty and Causality
- Belief state estimation
- Uncertainty modeling in dynamic systems with incomplete or noisy information
- Research in probabilistic modeling, causal inference, or dynamic systems
4. Multimodal Learning in Real Environments
- Vision-language models
- Learning from asynchronous, noisy, or partially missing modalities
- Sensor fusion or multimodal representation learning
- Publications or projects involving multimodal models
- Experience working with real-world (not solely synthetic) data
What You Will Work On
Your projects will be tailored to your expertise and may encompass:
- Designing AI systems that effectively navigate complex environments

