Video Platform Engineer

About IRL (Instawork Robotics Labs)

Researchers at UC Berkeley have identified a "100,000-year data gap"—the gulf between what trained AI language models and what physical robots actually have to learn from. Closing that gap is the defining infrastructure challenge of the physical AI era. IRL is Instawork's answer to it. We deploy skilled workers into real commercial and residential environments—kitchens, warehouses, hotel floors, and homes—to capture the high-fidelity task data that the world's leading robotics labs use to train their foundation models.

 About the Role

Every dataset IRL ships starts as raw video: multi-camera rigs recording a worker palletizing a warehouse order, cooking a meal, or turning over a hotel room. Many hours of synchronized streams per session, plus the telemetry captured alongside them, growing into petabytes.

As a Video Platform Engineer, you'll own the pipeline that turns that raw capture into training-ready data: ingest, transcode, time-synchronization, automate video quality checks, indexing, and delivery to the robotics labs that train on it.

This is a hybrid media-engineering and distributed-systems role. You'll make codec and container decisions. You'll design orchestration that survives a rig failing mid-session. And you'll own the cost per captured hour of processing.

You'll help shape the roadmap by working directly with the data collection teams in the field, and you'll have real say in which of their problems we solve first.

Who You Are

- 5+ years building and operating production video or media processing systems at scale.
- Hands-on, production experience with FFmpeg, GStreamer, or an equivalent media framework. You've built a pipeline around it and debugged it under load.
- Fluency in codecs and containers: H.264/HEVC/AV1, MP4/fMP4, and the practical tradeoffs between them.
- Strong understanding of distributed systems, cloud infrastructure, object storage, and observability.
- Strong programming and problem solving skills.
- Experience with durable workflow orchestration (Temporal, Conductor, Step Functions, Airflow, or similar).
- Experience designing cost-efficient architectures for large media or data workloads, and a working understanding of storage tiering, egress, capacity planning, and GPU utilization.
- Experience with Infrastructure as Code, CI/CD, and production operations.

**Preferred**

- Multi-camera, multi-view, or time-synchronized capture systems, and the clock-drift and alignment problems that come with them.
- Robotics or sensor data formats: MCAP, rosbag, ROS/ROS2, Parquet, or similar columnar and log formats.
- Familiarity with machine learning on video or images. Classification, detection and tracking, segmentation, re-identification, super-resolution, quality scoring, auto-annotation, redaction.
- Perceptual quality measurement (VMAF or equivalent) and encoding-ladder design.
- Streaming and packaging formats: HLS, DASH, CMAF, and DRM integration.
- Range across the stack. If you've built the annotation or review tooling on top of your own pipeline, we want to hear about it.

What You'll Do

- Design and build the ingest path from capture rigs to cloud storage, resilient to partial uploads, flaky field networks, and mid-session hardware failure.
- Build transcoding and normalization pipelines that turn heterogeneous rig output into consistent, analysis-ready media.
- Time-synchronize multi-camera video with telemetry and sensor streams into coherent, labeled episodes.
- Build automated quality checks that flag unusable sessions before they reach a customer - blur, exposure failure, dropped frames, etc.
- Design dataset versioning, lineage, and indexing so that any shipped episode can be traced back to its raw capture.
- Build the delivery path that gets datasets to robotics labs, including packaging, integrity verification, and access control.
- Own cost per captured hour and cost per delivered dataset: storage tiering and lifecycle, egress, encoding efficiency, and GPU utilization.
- Improve observability across the pipeline—monitoring, logging, tracing, and per-stage throughput and failure metrics.
- Troubleshoot complex issues spanning media tooling, storage, networking, and application code.
- Participate in an on-call rotation and own diagnosis and resolution of production issues.
- Participate in code reviews and raise the technical bar through clear, practical feedback.
- Own complex initiatives from problem definition through design, implementation, and operation.
- Measure and improve pipeline throughput, end-to-end latency from capture to delivery, dataset yield, reliability, incident recovery time, and infrastructure cost per workload.

Perks

- Free snacks
- Health Insurance
- Personal Insurance
- Flexible Hours
- Maternity/Paternity Leave
- Broadband Reimbursement

Compensation Band: 0.0 - 0.0