Principal Research Engineer – Agentic AI

Job Description

At Oracle Analytics, we are building the next generation of enterprise AI products to enable intelligent data analysis eraging our foundational strengths in data management and enterprise software applications, we are advancing our platforms and applications by deeply embedding cutting‑edge agentic AI, generative AI, and innovations in machine learning and optimization.

Our AI and Applied Science team is seeking a highly motivated Principal Research Engineer to perform innovation in agent planning, reasoning, and the efficient implementation of multi‑step reasoning agents across Oracle’s AI Data platform and intelligent applications. In this role, you will research, architect, and prototype next‑generation agent planning and orchestration techniques (hierarchical planning, tool routing, memory, verification/critique, tree/graph search) and efficiency methods (distillation, quantization, caching, speculative decoding, retrieval) to improve latency, cost, and reliability for enterprise workloads.

You will design and implement verifiable, auditable reasoning systems—spanning reasoning traces, intermediate state management, and outcome validation—to drive trustworthy outcomes in production. Partnering closely with applied scientists, research engineers, and product teams, you will take solutions from lab to production, delivering measurable impact in globally scaled intelligent applications.

This role requires deep expertise in agentic and generative AI with a strong focus on planning and multi‑step reasoning. Hands‑on proficiency with modern ML/LLM stacks (e.g., Python, PyTorch/JAX, transformers, vector stores, orchestration frameworks) and systems performance techniques (KV‑cache management, batching, routing, parallelization) is a must. Familiarity with post‑training/fine‑tuning (RLHF/RLAIF, DPO, PEFT), alignment and guardrails, and symbolic‑neural hybrid methods is encouraged to partner effectively with modeling and platform teams.

• Collaborate with applied scientists and research engineers to architect, develop, and evaluate agent planning and multi‑step reasoning solutions that meet enterprise requirements for accuracy, cost, and latency.
• Design and implement agent planning algorithms (hierarchical/task decomposition, tree‑of‑thought/graph search, self‑consistency/debate/critique, scheduler/executor patterns) and robust tool‑use orchestration with verification and fallback strategies.
• Build scalable, production‑grade inference and orchestration pipelines (retrieval, function calling, memory, streaming, caching) with strong observability, tracing, and telemetry for reasoning steps and tool interactions.
• Develop efficiency techniques (distillation/quantization, speculative decoding, adaptive routing, KV‑cache and batching strategies) to optimize throughput and stability under real‑world traffic.
• Instrument rigorous evaluation frameworks for reasoning quality and robustness.
• Partner with data engineers on data collection/curation/annotation guidelines for reasoning and tool‑use training signals (traces, plans, critiques, rewards).
• Work closely with product and platform teams to ship secure, compliant, and reliable capabilities into applications.
• Identify new opportunities for scientific exploration and evaluate emerging planning/reasoning technologies; communicate findings and best practices across teams.
• Maintain solid understanding of industry trends and developments in agentic AI, reasoning, and efficient LLM systems.

• PhD in Computer Science, Mathematics, Statistics, Engineering, or a related field.
• Experience in AI/Generative AI/Machine Learning in an industry setting, with emphasis on agent planning and multi‑step reasoning.
• Publication record in first‑tier venues (e.g., NeurIPS, ICLR, ICML, ACL, EMNLP, NAACL) is a plus.
• Hands‑on experience taking reasoning‑centric systems from prototype to production, including tool‑enabled agents, retrieval‑augmented reasoning, and verifiable execution with auditability and guardrails.
• Strong understanding of state‑of‑the‑art LLMs and agent frameworks, including planning/decomposition methods, tool routing,…