Senior Applied Scientist – 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. We are seeking a Senior Applied Scientist (with a PhD preferably) to perform innovation in AI agents for enterprise analytics—focusing on planning, multi‑step reasoning, and tool‑augmented execution with large language models over unstructured data (documents, logs, emails, multimodal artifacts).

You will design agent architectures, memory and planning systems, retrieval and grounding pipelines, and evaluation frameworks that deliver reliable, auditable, and cost‑efficient enterprise outcomes. You will partner closely with research engineers and product teams to ship agentic systems to production, rigorously evaluate reasoning quality and safety, and drive measurable customer and business impact.

• Perform end‑to‑end agentic system development: define agent goals and decomposition strategies; design planners, controllers, and executors; implement tool‑use orchestration (APIs, SQL, vector search, code execution) and robust recovery/rollback.

• Advance planning and reasoning: hierarchical/task planning, self‑reflection and critique, debate/tree‑search methods, constraint satisfaction, and chain‑of‑thought/tool former‑style approaches to improve correctness, faithfulness, and robustness.

• Ground LLMs on unstructured data: build retrieval and indexing over documents, semi‑structured data.

• Ensure safety, privacy, and compliance: content safety policies, least‑privilege tool access, execution sandboxes, prompt/memory redaction, PII handling, and governance appropriate for regulated enterprise settings; implement interpretable action logs.

• Productionize agentic solutions: collaborate with platform teams to ship planning/orchestration services and evaluation harnesses; implement observability, telemetry, canarying, rollback, and lifecycle management for agent workflows.

• Stay current with research and translate advances into production differentiators; mentor teammates and contribute to a culture of scientific rigor and impact.

• PhD in Computer Science, Machine Learning, Statistics, Electrical Engineering, or related field with a focus relevant to LLMs, planning/reasoning, NLP, or autonomous/interactive systems.

• Experience (industry or applied research) building and deploying ML/LLM systems, including agentic workflows, retrieval/grounding, and evaluation at scale.

• Demonstrated expertise in agentic methods: multi‑step planning, tool‑use orchestration, reflection/critique, and structured reasoning (e.g., CoT, programmatic planning).

• Strong background in retrieval over unstructured data, RAG architectures, document preprocessing, indexing, and provenance tracking for accuracy, safety, and robustness.

• Proficient in Python and modern ML stacks:
PyTorch/JAX, Transformers, vector databases/IR libraries, orchestration frameworks; solid software engineering practices and experimentation discipline.

• Track record of publications in top venues (e.g., NeurIPS, ICML, ICLR, ACL, EMNLP, NAACL) or equivalent demonstrated impact in production systems.

• Experience designing data/feedback pipelines for agent evaluation: step‑level labeling, trace audits, and active learning for hard cases; familiarity with bias/variance trade‑offs.

• Knowledge of search/planning and decision‑making: tree search, bandits for tool/model selection, off‑policy evaluation for policy changes, and statistical testing for online experiments.

• Familiarity with LLM efficiency and serving: PEFT/LoRA/QLoRA, quantization, KV cache management, batching, speculative decoding, routing across models/skills, and throughput/latency trade‑offs.

• Experience integrating safety/guardrails and policy enforcement: sandboxed tool execution, OAuth/secret management, rate‑limiting, jailbreak/prompt‑injection…