Job Description

About us

At Encord, we're building the AI infrastructure of the future. The biggest challenge AI companies face today is not nearly as glamorous as the outside world may think: it's all about data quality. In fact, the success of any AI application today relies on the quality of a model's training data. For 95% of teams, this essential step is both the most costly, and the most time-consuming in getting their product to market.

As ex-computer scientists, physicists, and quants, we felt first-hand how the lack of tools to prepare quality training data was impeding the progress of building AI. AI today is what the early days of computing or the internet were like, where the potential of the technology is clear, but the tools and processes surrounding it are still primitive, preventing the next generation of applications. This is why we started Encord.

We are a team of 100+, working at the cutting edge of multimodal and visual AI. Encord is backed by top investors, including CRV, Y Combinator, and Next47, leading industry executives like Luc Vincent, former VP of AI at Meta, and other Bay Area AI leaders. We are one the fastest growing companies in our space, and consistently rated as the best product in the market by our customers.

The Role

We're building a robotics data collection operation from scratch — and we need someone to figure out how to make it work. There's no playbook. No inherited infrastructure. No one to tell you what to do next.

You'll own the entire stack: robot arms, sensors, VR teleoperation rigs, data pipelines, quality systems, and customer delivery. We currently operate leader-follower teleoperation setups and are actively building toward VR-based interfaces for higher-fidelity remote collection. You'll help us make that transition and define what the pipeline looks like at the other end.

This is a 0-to-1 role. If you need clear instructions and a defined scope, this isn't for you. If you get energized by building something from nothing, keep reading.

What You'll Do

• Build the hardware infrastructure — Set up and maintain robot arms, cameras, and VR teleoperation rigs. Design workstation layouts. Keep systems calibrated. When something breaks mid-shift, get us back online. When we scale from 10 to 50 stations, figure out how.
• Build the data pipeline — Ingest multi-modal sensor streams (joint states, wrist and scene cameras, gripper, 6DOF controller poses), synchronize across modalities, validate quality, and export in formats customers need — LeRobot HDF5, RLDS, MCAP, ROSBAGS and custom specs. Own the full journey from sensor to deliverable dataset.
• Maintain edge infrastructure — Run local buffering and automated upload at capture facilities. Keep data flowing reliably from the floor to the cloud even when connectivity is flaky.
• Operationalize annotation tooling — Work with our ML team to run VLM/VLA-based annotation passes that auto-generate action labels from raw robot video. Own throughput, prompt reliability, and output validation.
• Solve problems as they come — Debug hardware failures, adapt to new customer requirements, work around sensors that don't behave as expected. Document what you learn so we don't hit the same wall twice.
• Shape what comes next — Help us figure out when to specialize, what to build vs. buy, and how to scale globally.

You Might Be a Fit If...

• You've built something from scratch before — a lab setup, a data collection system, a side project, a startup — and you loved the ambiguity
• You default to action. When you don't know the answer, you run an experiment instead of waiting for direction
• You have hands-on experience with robotics, automation, or mechatronic systems — robot arms, drones, CNC machines, or something you built yourself
• You're proficient in Python and comfortable building data pipelines from scratch
• You're at home in Linux and ROS/ROS 2, and can pick up new tools quickly (Docker, cloud infra, whatever's needed)
• You understand what makes a robot episode good training data — not just that it recorded, but that it's actually usable for imitation learning
• You've dealt with multi-modal timestamp synchronization — hardware triggers, PTP, or software alignment across cameras and joint encoders
• You'd rather ship something imperfect and iterate than wait for perfect requirements
• You take ownership end-to-end, even when it's outside your "job description"

Bonus Points

• Experience with VR-based teleoperation (e.g., Quest + ROS bridge, custom motion or haptic interfaces)
• Familiarity with VLA training pipelines — OpenVLA, π0, GR00T N1, SmolVLA, or similar
• You've worked with MCAP, rosbags, or other robotics data container formats
• You've shipped something in a previous startup or early-stage environment
• You have a GitHub, blog, or portfolio that shows what you've built

What We Offer

• Competitive salary and equity in a hyper growth startup.
• Real opportunities to grow. We’re growing insanely fast and you’ll have all the opportunities for growth that you can handle.
• Strong in person culture. We're 3-5 days a week in our newly-launched loft office in North Beach.
• Flexible PTO to recharge.
• Annual learning and development budget.
• 18 paid vacation days + federal holidays.
• Lots of opportunities for travel (all around the US, to London & Europe).
• Bi-annual off-sites and monthly socials.
• Health, dental and vision insurance.

Encord offers a unique opportunity to be part of a startup with a clear mission and vision. You will get to explore and build services enterprise AI use cases across many different industry verticals such as healthcare, surveillance, retail, agriculture, and many more.

Our work is at the cutting edge of computer vision and deep learning, which also includes working on solving unsolved problems within those fields.

We may use artificial intelligence (AI) tools to support parts of the hiring process, such as reviewing applications, analyzing resumes, or assessing responses. These tools assist our recruitment team but do not replace human judgment. Final hiring decisions are ultimately made by humans. If you would like more information about how your data is processed, please contact us.