Member of Technical Staff - ML Data Infrastructure

Mimetic Machines; München; On-site; EUR 90,000 - 120,000; Full time

Job summary

Mimetic Machines is hiring to own the data platform for real-world recordings, from capture to training-ready data and delivery. You will design and operate robust ingestion, processing, and storage pipelines that tolerate interruptions and network issues.

You will work closely with researchers and the founders to define architecture, implement scalable systems, and ensure data quality and provenance as the dataset grows. This is a hands-on founding role with immediate impact.

Qualifications

• You’ve owned a production data pipeline for video or sensor data and can explain decisions and trade-offs.
• You understand object storage, asynchronous processing, retries, and failure modes.

Responsibilities

• Own the data platform from capture to training-ready datasets and delivery.
• Build reliable recording, offload, and upload workflows that recover from interruptions and unreliable networks.
• Design ingestion contracts for different camera configurations and modalities.
• Build processing pipelines for validation, reconstruction, annotation, and other data transformations with reliable scheduling and retry logic.
• Make datasets reproducible through versioning and provenance across sources, processing, annotations, and exports.
• Collaborate with researchers to produce useful dataset formats and training inputs.
• Make pipeline health and data loss visible; fix problems at their source with field operations.

Skills

Data platform ownership

Reliable recording pipelines

System resilience

Cost & throughput monitoring

Independent progress

Tools

Object storage

Asynchronous processing

Retries & recovery

Workflow orchestration

Cloud infrastructure

GPU processing

Annotation tools

Job description

Own the systems from recording and upload through storage, processing, quality review, and delivery as training-ready data.

About Mimetic Machines

Mimetic Machines builds learning and deployment environments for physical AI. We capture skilled work in real workplaces and build the data, tools, and environments that help robots learn to do useful work.

The role

We’re collecting a growing dataset of unique real-world physical work. You’ll own the systems that make that collection useful: from recording and upload through storage, processing, quality review, and delivery as training-ready data.

The recordings come from different environments, camera configurations, and sensor types. Connections fail. Sessions get interrupted. Processing jobs need to recover. Researchers need to know exactly what a dataset contains and how it was produced.

You’ll take responsibility for that entire path, working directly with the founders. That means setting the architecture, building the systems, and making sure they hold up as collection grows. Your work will determine how quickly we can turn a recording in the field into something useful for research.

What you’ll do

• Own the data platform from capture to training-ready datasets and delivery.
• Build reliable recording, offload, and upload workflows that recover from interrupted sessions and unreliable networks.
• Design ingestion contracts for different camera configurations and modalities.
• Build processing pipelines for validation, reconstruction, annotation, and other data transformations, with reliable scheduling, retries, and recovery.
• Make datasets reproducible through versioning and provenance across source recordings, processing outputs, annotations, and exports.
• Work with researchers to produce useful dataset formats, subsets, and training inputs.
• Make pipeline health, processing cost, and data loss visible. Work with capture and field operations to fix problems at their source.

What we look for

• You’ve owned a production pipeline for video, media, sensor data, or large ML datasets, and can explain the decisions and trade-offs you made.
• You understand object storage, asynchronous processing, retries, and what happens when a system fails partway through.
• You measure correctness, cost, and throughput, and use those measurements to decide what to improve.
• You can turn an ambiguous objective into a concrete plan and make iterative progress independently.
• You take responsibility for the outcome, from the first implementation through operating and improving it.

Experience that could be useful

Video encoding and compression, multi-camera or multimodal data, dataset formats and sharding, GPU processing, annotation tools, cloud infrastructure, workflow orchestration, or systems programming.

You don’t need experience in every area. We value strong engineering judgment and evidence that you can learn, build, and keep systems running.

Working at Mimetic

This is a founding role with room to shape both the platform and how we work. You’ll stay close to the people collecting the data and the researchers using it. You’ll see where the process breaks and have the ownership to improve it.

We want someone who enjoys taking a complicated, unfinished system and making it dependable.