Engineer In Reasoning, Planning, and Decision-making systems

Airbnb

17d•$296,000 - $370,000•Remote

About The Position

Airbnb is seeking a Principal / Distinguished AI/ML Researcher and/or Engineer with deep experience in reasoning, planning, and decision-making systems. This role is ideal for individuals who have architected post-training intelligence frameworks, integrated Large Reasoning Models (LRMs) with Knowledge Graphs, and applied Reinforcement Learning (RL) as a first-class component of adaptive planning and control. You will be responsible for inventing, scaling, and operationalizing intelligent decisioning substrates that blend symbolic and sub-symbolic methods, enabling next-generation AI systems that go beyond pattern recognition into the realm of deliberation, foresight, and agency. Our mission is to build cognitive AI systems that combine post-trained foundational models, explicit memory and knowledge, and recursive planning strategies to power sophisticated real-world decisioning in personalized environments. You will collaborate across disciplines and influence company-wide AI architecture. A core dimension of this role is the design and deployment of multi-agent systems, where reasoning, planning, and decisioning are distributed across networks of intelligent agents. You will formulate coherent, synergistic strategies that enable agents to cooperate, negotiate, and align objectives, ensuring that distributed intelligence converges to purposeful, high-quality outcomes across contexts.

Requirements

Masters or equivalent in Computer Science, AI, Cognitive Science, or related fields.
Recent published work or patents in AI, Cognitive Science, or related fields.
15+ years in AI/ML, including post-training architectures and production-scale reasoning systems.
Advanced coding proficiency in Java, Python, C++, or similar, with experience in ML/RL frameworks (e.g., PyTorch, Ray, JAX, RLlib) at scale.
Proven experience integrating LLMs/LRMs with Knowledge Graphs or structured world models.
Deep understanding of Reinforcement Learning and its application to decisioning and planning.
Fluency in hybrid model architectures: connectionist-symbolic fusion, retrieval-based agents, or goal-directed transformers.
Experience working on multi-agent coordination, distributed RL, or cooperative inference systems.

Nice To Haves

Ph.D. in AI, Machine Learning, Robotics, Cognitive Systems, or related areas.
Published work or patents in multi-agent reasoning, plan synthesis, knowledge-augmented learning, or generative control.
Experience in cognitive architectures, neuro-symbolic systems, or agent-based simulation environments.
Demonstrated ability to lead cross-functional research-to-production transitions.
Experience with memory architectures, task graphs, or semantic program induction.
Prior work on distributed intelligence platforms with explicit agent interaction models and collective decision-making logic.

Responsibilities

Drive foundational and applied research in reasoning engines, planning architectures, and decision-making frameworks at scale in order to incorporate genAI into the ranking / recommendation / personalization stack in both single model to multi-agent ( system ) level intelligence with objective to grow the business (new user growth, abandoned user, long tailed user) in existing and new business areas while supporting Multi-Modal NL → Conversational Interfaces.
Advance techniques in LLM/LRM post-training, reinforcement learning–based decisioning, and knowledge-integrated agents.
Design methods for plan induction, value estimation, and contingency modeling within intelligent agents.
Explore and validate protocols for distributed reasoning and joint planning among cooperative agents in multi-agent systems.
Architect RPD systems that integrate post-trained LLMs/LRMs, graph-structured memory (e.g., KGs), and RL-driven controllers.
Design recursive task planners, search-based or policy-based reasoners, and belief-state trackers that can interoperate with large model substrates.
Ensure modularity and extensibility through multi-agent frameworks, agentic substrates, and declarative planning pipelines.
Define communication protocols, coordination strategies, and cross-agent knowledge alignment mechanisms to foster emergent cooperative intelligence.
Build and evolve stateful, dynamic models that combine supervised learning with online/offline reinforcement, simulation-based rollouts, and symbol grounding.
Implement hybrid pipelines that couple learned embeddings, prompted generative models, and graph-theoretic inference.
Optimize systems for adaptive exploration, planning horizon control, and policy robustness.
Develop frameworks for distributed value propagation, multi-agent credit assignment, and global planning from local agents.
Set direction for planning/reasoning infrastructure within the AI/ML platform strategy.
Serve as the technical conscience and architectural leader across high-stakes AI initiatives involving autonomous agents or high-fidelity decision pipelines.
Mentor teams in systems thinking, causal modeling, symbolic-connectionist integrations, and long-term planning under uncertainty.
Lead development of multi-agent reasoning systems, defining principles for inter-agent knowledge exchange, goal delegation, and cooperative decision resolution.
Work across disciplines—product, infra, and design—to translate ambiguous product intent into multi-stage reasoning pipelines.
Partner with researchers, ontologists, and ML engineers to encode world knowledge, goals, and values into usable inference artifacts.
Contribute to a company-wide understanding of what it means to make intelligent choices, not just predictions.
Collaborate with internal teams on distributed agent coordination, shared memory protocols, and policy harmonization across decision surfaces.
Productionize real-time reasoning loops with low-latency inference, caching, retrieval-augmented generation, and streaming updates to symbolic memory.
Deploy post-training hooks for inserting logic, constraints, and domain priors into existing large models.
Create advanced monitoring, attribution, and evaluation pipelines for agent behavior and decision quality.
Operationalize multi-agent orchestration, ensuring reliable and fault-tolerant communication and decision propagation.