Software Engineer, Perception Attributes Autolabeling Pipeline in , United States

The Perception Attribute Flywheel team is looking for a Software Engineer to build and operate the autolabeling pipeline that accelerates human annotation throughput on vehicle attribute classification tasks. The company is building a future for Riders, not drivers. The accuracy of our perception attribute models - recognizing emergency vehicles, school buses, brake lights, hazard signals, and more - depends on a steady flow of high-quality labeled examples drawn from our fleet's drive data. Today, every label is produced by a human annotator from scratch. We are building a pipeline that uses off-the-shelf foundation models (Gemini, SigLIP, CLIP) to pre-label tasks, so human reviewers verify and correct rather than labeling from scratch.

This role owns the pipeline engineering for that system: ingesting queued tasks from our annotator service, calling foundation-model APIs at fleet scale, writing structured predictions back into the labeling workflow, and operating the whole thing reliably. The team lead and supporting ML engineers own model selection, prompt design, and evaluation methodology; this role partners closely with them but is not expected to own those decisions. If you take pride in building reliable, observable, well-tested data pipelines and want to ship a system that visibly accelerates an autonomous vehicle program, you will excel in this role.

Responsibilities:

Build the autolabeling pipeline: ingest queued tasks from the annotator service, dispatch them to foundation-model APIs (Gemini and others), parse structured outputs, and write pre-labels back to the labeling workflow. Build the observability layer: per-task latency, per-model cost, per-attribute coverage, and error-mode dashboards. Run experiments designed by the team lead - set up the inputs, execute, and collect outputs in formats the ML engineers can analyze. Integrate the pipeline cleanly with existing systems, partnering with the data infrastructure team. Document the system, write runbooks, and ensure a clean handoff at end of the engagement., End-to-end ML pipeline stewardship - owned an ML system in production from data ingest through inference through monitoring. Annotation tooling or human-in-the-loop ML workflows. Autonomous-systems data pipelines. AWS, especially S3, ECS/EKS, Lambda. Working in a codebase shared with ML engineers (proto schemas, joint deploys).

3+ years of backend/data pipeline engineering experience; Strong Python; comfort with C++; Large-dataset experience with PySpark or equivalent. ML fundamentals - understanding of model inference, embeddings, structured output, and common eval metrics (precision, recall, calibration); able to reason about ML data shapes and integration patterns. Experience integrating foundation models (Gemini, OpenAI, Anthropic) at production scale; Excellent written communication for design docs and runbooks., 3+ years of backend/data pipeline engineering experience. Strong Python; comfort with C++. Large-dataset experience with PySpark or equivalent. ML fundamentals - understanding of model inference, embeddings, structured output, and common eval metrics (precision, recall, calibration); able to reason about ML data shapes and integration patterns. Experience integrating foundation models (Gemini, OpenAI, Anthropic) at production scale. Excellent written communication for design docs and runbooks.