ML and AI Knowledge Systems Engineer

Advanced Micro Devices, Inc•San Jose, CA

1d•Hybrid

About The Position

We are building GoldenEye, an internal AI knowledge platform that lets engineers ask natural-language questions across GPU design knowledge, including RTL, microarchitecture specifications, verification collateral, Confluence, and JIRA, and receive trustworthy answers with citations. A prototype is already used by hardware and software teams. We are now scaling it to thousands of users, with the potential to serve more than 40,000 engineers worldwide. We are seeking a hands-on PMTS-level ML engineer to lead GoldenEye’s retrieval, ranking, and answer-quality architecture. You will turn a promising prototype into an authoritative production platform while setting technical direction, mentoring engineers, and partnering with hardware, software, infrastructure, and security teams. GoldenEye will make critical engineering knowledge easier to find, verify, and use, accelerating GPU design, verification, debugging, and bring-up. You will serve as the technical anchor for its ML core and shape its adoption across the engineering organization.

Requirements

Deep relevant experience building production ML, search, ranking, or information-retrieval systems.
Hands-on experience with RAG, embeddings, vector and keyword search, reranking, chunking, and retrieval evaluation.
Practical experience with LLM prompting, structured generation, tool calling, evaluation, and hallucination reduction.
Experience designing ML evaluation datasets, metrics, experiments, and regression tests.
Demonstrated ability to translate current ML or information-retrieval research into rigorous experiments and production improvements.
Strong software and systems engineering skills, including APIs, asynchronous processing, observability, and production operations.
Experience leading architecture, mentoring engineers, and influencing senior technical stakeholders.
Experience with distributed inference systems such as vLLM and GPU performance optimization.
Experience with AMD Instinct GPUs or comparable accelerators.
Experience with vector databases, hybrid search, distributed indexing, or authorization-aware retrieval.
Familiarity with MCP, coding agents, or tool-based agent architectures.
Experience with hardware design, EDA, RTL, microarchitecture, firmware, compilers, or chip verification.
Technical thought leadership demonstrated through publications, patents, open-source contributions, conference participation, or influential production systems. Publications are valued but not required.

Nice To Haves

Master’s or Doctoral degree in machine learning, information retrieval, NLP, computer science, electrical or computer engineering, or a related field.

Responsibilities

Own the architecture for embeddings, chunking, hybrid retrieval, reranking, multi-query planning, and reciprocal-rank fusion.
Build retrieval workflows across specifications, RTL, verification artifacts, wikis, and issue trackers.
Improve search for domain-specific content such as ISA mnemonics, registers, signal names, acronyms, and code identifiers.
Reduce hallucinations through grounded generation, citation enforcement, source validation, and conflict resolution.
Define source-authority and freshness policies for conflicting or outdated information.
Build an evaluation framework for retrieval relevance, answer correctness, citation faithfulness, latency, cost, and regressions.
Use expert feedback, production data, hard negatives, and targeted failure cases to create representative evaluation datasets.
Track advances in retrieval, RAG, agentic systems, evaluation, and efficient inference, and evaluate promising techniques against production requirements.
Optimize embedding, reranking, and inference workloads on AMD Instinct GPUs, including multi-GPU and multi-node serving.
Partner with platform teams on scalable services, indexing, data reconciliation, observability, and cluster orchestration.
Advance Model Context Protocol (MCP) tools used by coding agents.
Ensure retrieval, caching, and answer generation respect per-user access controls.
Mentor engineers, review designs, and communicate technical decisions across organizations.