Software Engineer, AI Platform

What matters most

• Discover the right product with customers

  Role: Work directly with customers, engineers, and product leaders to identify useful capabilities in an ambiguous, zero-to-one role. Preparation: Be ready to describe how you would turn an observed engineering workflow into a prototype and use feedback to decide what to build next.

• Build the AI platform foundations

  Role: Build agent infrastructure, evaluation pipelines, retrieval systems, and tools for reasoning over engineering data. Preparation: Be ready to explain design choices and how you would evaluate whether these components help engineers.

• Ship investigation workflows

  Role: Deliver AI-powered workflows for anomaly investigation, root cause analysis, and understanding complex hardware behavior. Preparation: Think through how an engineer would move from an observed anomaly to evidence they can inspect.

• Make data-intensive systems production-ready

  Role: Build robust systems across large-scale time series, telemetry, logs, and other engineering datasets; the posting calls for scalable backend or distributed-systems fundamentals. Preparation: Have a concrete approach to data access, scale, and operational reliability.

• Set patterns as the team scales

  Role: Help establish engineering patterns, infrastructure, and best practices for Hardware Intelligence. Preparation: Consider which prototype decisions you would revisit before making them shared foundations.

Likely challenges to think through

• Balance fast experiments with durable foundations

  Likely challenge, not a confirmed problem: Rapid customer-tested prototypes may compete with the need for production-ready systems and reusable infrastructure. Prepare to explain when you would keep an approach provisional, harden it, or retire it.

• Make AI investigations useful without overstating conclusions

  Likely challenge, not a confirmed problem: An anomaly or root-cause workflow must help engineers reason over evidence, while an AI-generated explanation could appear more certain than the data supports. Prepare to discuss evaluation and ways to expose supporting evidence and uncertainty.

• Connect varied engineering data at scale

  Likely challenge, not a confirmed problem: Reasoning across large-scale time series, telemetry, and logs may require trade-offs between retrieval quality, responsiveness, and system complexity. Prepare to describe how you would test those trade-offs against a specific investigation workflow.

Topics to prepare

• Time-series and telemetry data design

  Review timestamp alignment, missing data, event correlation, and storage/query trade-offs. Exercise: sketch a data model that connects sensor readings and logs during a failure window. The role explicitly works with large-scale time series, telemetry, and engineering datasets.

• Scalable backend and distributed-system trade-offs

  Review how ingestion, processing, and investigation queries behave under scale and partial failure. Exercise: diagram a backend for engineering-data investigations and explain its bottlenecks and recovery paths. Scalable backend or distributed-system design is explicitly sought.

• Retrieval, agents, and evaluation

  Review how retrieved evidence can support an agent’s conclusions and how to measure incorrect or unsupported outputs. Exercise: define test cases and success criteria for an agent investigating an anomaly. Agent infrastructure, retrieval systems, and evaluation pipelines are named responsibilities; LLM applications are listed as nice to have.

• Anomaly investigation and root-cause reasoning

  Review the distinction between a detected anomaly, correlated signals, and a supported root-cause hypothesis. Exercise: map a hypothetical hardware failure from observations to competing explanations and evidence needed to distinguish them. These investigations are explicit product workflows.

• Zero-to-one product decisions

  Review how to turn an undefined customer problem into a testable capability without committing prematurely to a full system. Exercise: draft a short problem statement, prototype scope, and feedback criteria for an engineering investigation workflow. Customer discovery, rapid prototyping, and iteration are explicit parts of the role.

Skills to revise

• Investigate a telemetry failure window

  Using a small practice dataset, align time-series readings with logs, identify an unusual interval, and record what the evidence does and does not establish. This rehearses the posting’s hands-on work with telemetry, anomalies, and root-cause analysis; the practice dataset is a preparation suggestion.

• Prototype evidence retrieval

  Build a small practice workflow that retrieves relevant log excerpts and telemetry intervals for an investigation question, then checks whether the returned evidence matches the time window. Retrieval over engineering data is an explicit responsibility; this prototype is a suggested exercise.

• Construct an evaluation set

  Write several investigation cases with expected evidence, plausible wrong conclusions, and a scoring rubric; run a prototype against them and inspect failures. This practices the evaluation pipelines named in the posting.

• Design a production-readiness review

  Take a prototype investigation service and document its scaling limits, failure handling, data-quality checks, and observable behavior. The exercise reflects the explicit expectation to turn approaches into robust, production-ready systems and establish engineering patterns.

• Translate customer feedback into an iteration

  Role-play a conversation with a hardware engineer about a difficult investigation; summarize the current workflow, prototype one improvement, and specify what feedback would change the next version. Direct customer engagement and feedback-driven iteration are explicit responsibilities; the role-play is a preparation suggestion.

• Prepare a production AI project walkthrough

  The role calls for AI or ML products shipped to production. If applicable, prepare one example covering your contribution, architecture, deployment decisions, measured results, and lessons learned.

• Sketch a telemetry investigation system

  As a practice exercise, draw a backend design for ingesting large-scale time series and logs and retrieving data for anomaly investigation. Be ready to explain scaling, data quality, and operational tradeoffs.

• Outline an AI workflow and its evaluation

  For a hypothetical root-cause investigation, show where retrieval, agent tools, and human review would fit. Propose tests for whether findings are useful and supported by engineering data; these are preparation ideas, not specified product requirements.

• Document a zero-to-one decision

  If applicable, prepare an example of an undefined customer problem you helped turn into a prototype and then a product decision. Identify the feedback gathered, the technical bet made, and what changed after testing.

• Gather evidence of engineering judgment

  Select examples, where available, of scalable backend or distributed-system work and handling large datasets. Prepare concrete tradeoffs and outcomes rather than a list of technologies.

• Write questions for the team

  Ask how Hardware Intelligence users currently investigate failures, how prototypes are evaluated with customers, and which engineering patterns the team most needs to establish as it scales.

• Prepare factual answers to role logistics

  The posting specifies five days per week in an office in New York, Los Angeles, Austin, or Seattle and U.S. export-control requirements. Be ready to answer the employer’s location and authorization questions accurately, without assuming eligibility.