Staff Software Engineer, ML Data Infrastructure, Autonomy

Rivian•Palo Alto, CA

About The Position

Rivian's Autonomy org needs a Staff Software Engineer, ML Data Infrastructure to own how autonomy data is described, indexed and accessed. This sits in the Platform Services team in the AI Platform organization in the Autonomy team. Our fleet of 100,000+ vehicles produces a continuous stream of multi-modal drive data, plus the output of every model run against it and every simulation executed on it. The role requires deep expertise in columnar and analytical data stores, schema and format design, and the query and access layers that ML and analytics teams depend on. You'll work with the AI Platform, Perception, Planning, Simulation, and Vehicle Integration, Product Management, and other technology partners. Autonomy data currently lives across multiple systems because no single store serves all our access patterns: fleet-scale byte storage, petabyte-scale analytics, sub-second fleet-wide search, and small transactional state each have different cost and latency profiles. You'll own the unified metadata and query layer, decide what consolidates and what stays specialized, and land the migrations onto a unified data lake.

Requirements

Bachelor's degree in Computer Science, Electrical Engineering or a related field, or equivalent experience.
8+ years in software engineering, with a strong focus on data infrastructure or data platform work.
5+ years owning a large-scale data platform for ML or analytics workloads, including its storage and access model, not only the pipelines on top of it.
5+ years with columnar and analytical stores (ClickHouse, Pinot, Druid, BigQuery, Databricks or Snowflake), including data modeling for high-cardinality data and query performance tuning.
3+ years designing storage and query layers for large volumes of heterogeneous metrics, such as KPIs or performance telemetry, where diversity is as hard a problem as volume.
3+ years with document or NoSQL stores: schema design, indexing strategy, sharding and the operational realities at scale.
3+ years with modern data and table formats (Parquet, Arrow, Iceberg, Delta), with clear reasoning about when each applies.
3+ years building batch and streaming pipelines at scale with production orchestration and real data quality gates.
3+ years hands-on with Python, plus production experience in Go, C++ or Rust for performance-sensitive parts of an access layer.
3+ years with queueing and event-driven systems (SQS, Kafka, Kinesis or equivalent) in a production ingest path.
2+ years with production monitoring and alerting for data pipelines (Prometheus, Grafana, Datadog or CloudWatch), including automated integrity and freshness checks.
2+ years migrating a production data platform between storage or format architectures incrementally, while it stayed in service.
Engineering leadership: setting technical vision, timelines and priorities for a project or team, acting as technical lead, and mentoring engineers.
Ability to turn ambiguous, high-level requirements into a detailed system design and drive it to completion unprompted.
Technical excellence: willingness to work through implementation detail, and a record of raising technical standards across a broader engineering organization.
A record of designing schemas and interfaces that other teams built on and that survived changing requirements.

Nice To Haves

Applied ML for data problems, such as data mining, active learning or embedding-based retrieval for rare and long-tail scenarios.
Robotics or AV data experience, including ROS/ROS 2, MCAP, rosbag and time synchronization across sensor modalities.
Vector databases and embedding-based retrieval for data mining and scenario discovery.
ML training data pipeline performance, including dataloader throughput, sharding and shuffling.
Feature stores, data catalogs or lineage systems.

Responsibilities

Own the unified data layer for autonomy: a single interface over document metadata, high-cardinality columnar analytics, real-time search, and object-stored sensor payloads, so engineers query concepts rather than databases.
Design the architecture for the datalake, metadata access and the APIs that expose it, giving engineers a single access layer
Own the data access and format strategy, including the canonical log format used in production today and the choice of ML-native columnar and vector access going forward. Define the canonical schemas, own their evolution, and drive the migration path.
Design and build batch and streaming pipelines for eval data, and the storage layer behind them, to handle both the volume and diversity of metrics from on-road and simulation runs at their actual cardinality.
Build the indexing and discovery layer for fast semantic and metadata search across the fleet's data: scenario tagging, event indexing, embedding-based similarity search, and the query surface engineers use.
Improve the data mining tools that apply ML techniques to data discovery, so Perception, Behavior and Planning engineers can find rare and long-tail scenarios at fleet scale rather than searching by hand.
Build well-documented tools and APIs so engineers outside data infrastructure can find, slice and materialize the data they need without writing a pipeline.
Treat data correctness as a discipline: schema validation, contracts between producers and consumers, freshness and completeness monitoring, and alerting that catches bad data before a model trains on it.
Build continuous testing and monitoring for the platform, covering ingest health, data freshness, schema conformance and query performance.
Design to the workflows of Perception, Behavior, Planning and Simulation engineers.
Work with the security & privacy team on retention, access control, consent handling and regional data requirements for vehicle-collected data.
Set data engineering standards across Autonomy and mentor engineers on schema design, query performance and pipeline reliability.