Founding MTS, Data & Data Infra

MetAntz; San Jose (CA); On-site; USD 180,000 - 260,000; Full time

Benefits offered by this job

Equity package

Health insurance

Flexible PTO

Remote work options

Latest hardware and compute resources

Job summary

PhyAgents in San Jose, CA seeks a Lead Data to own data strategy, data collection, and data infrastructure for a multimodal robotics platform. You will define data requirements, oversee remote data operations teams, and shape the tooling and processes that ensure high-quality, scalable data for model training.

You will drive end-to-end data ownership, build the data engineering stack, and partner with modeling and product teams to enable reliable, long-horizon robotic performance in industrial

Qualifications

• Defining data requirements for every model we train, including tasks, conditions, volume, and quality bar.
• Closing the loop from evals and deployments back into data collection cycles.
• Diagnosing whether a failed policy is a data problem and owning the answer.

Responsibilities

• Stand up the remote data ops team: hiring plan, onboarding, certification, and daily rhythm.
• Designing collection protocols: teleoperation, handheld, ego video, and operator actions per take.
• Designing the annotation scheme: granularity, taxonomy, and adjudication rules.
• Building tooling for data capture, validation, upload, and fast feedback on bad takes.

Skills

Python

Systems programming

People management

Data quality

Data infrastructure

Robotics data

Education

Ph.D./M.S./B.S. in Computer Science or related field

Tools

S3 / data lake

Ray

Spark

Parquet

Iceberg

ROS

Job description

"

Lead Data

About The Company

At PhyAgents, we are building the Claude Code for Physical AI — a robotic foundation model plus a physical agentic framework for reliable, long-horizon operations in industrial environments. We deliver complete robots as a workforce. We are starting with a small number of high-growth verticals where the work is mission-critical, the market is enormous, and fixed automation has never worked.

The founders are Stanford alumni: a serial entrepreneur and ex-VP at Intel who previously built and exited a robotics company, and a key foundation models lead from Apple. Our founding team includes executives and senior engineering leaders from top technology companies. We are seed-funded and backed by multiple Stanford professors and deep tech investors who run AI and compute at the top of the field. Joining now means owning a foundational piece of the stack, working directly with the founders, and shaping a company at the intersection of the two biggest platform shifts of our time, foundation models and robotics.

About The Role

This is the lead data role. Data is the moat, and you own it end-to-end. This role has three parts:

• Deciding what data we need, and delivering it. You work directly with modeling to define what gets collected — which tasks, which conditions, how many episodes, and what separates a usable episode from a discarded one. Then you go get it. You are the DRI for data — the quality bar is yours, and so is the delivery.


• Bringing up and running a remote data operations team. We collect and annotate our own data - teleoperation, UMI-style handheld collection, and ego video - and are building an overseas data ops team. You design what gets collected and how, and what gets annotated and how. You build, buy the tooling the team works in, write the process, train them on it, and then push throughput and quality up week over week, permanently. This part of the job is as much process design, diligence about data, and working with people across time zones as it is engineering.


• Building the infrastructure underneath it. Petabytes of multimodal data. Storage, data management, indexing, retrieval, search and versioning. You decide what to build and what to buy, and you own and maintain whatever you choose.

This is an early stage startup and a founding role. You own a big chunk of the stack end-to-end. There is no playbook to inherit and your decisions will set the technical trajectory of the company. Be an owner, visionary and a leader!

What You'll Own

Data strategy, with modeling

• Defining the data requirements for every model we train — tasks, conditions, modality mix, volume, and the explicit quality bar for each


• Closing the loop from eval and deployment failures back into the next collection run


• Diagnosing whether a stuck policy is a data problem, and being accountable for the answer


• Prioritizing across competing modeling asks when collection capacity is finite, and saying no

Data operations — tooling, process, and people

• Standing up the remote data ops team — hiring plan input, training material, onboarding, certification, and the day-to‑day operating rhythm


• Designing the collection protocols themselves — teleoperation, UMI‑style handheld, and ego video — including rig setup, scene and task variation, and what an operator is supposed to do on every take


• Designing the annotation scheme — what gets segmented and labeled, at what granularity, with what taxonomy, and how disagreements get adjudicated


• Building the tooling both sides live in — session capture, on-device validation, upload, review queues, annotation UI, and fast feedback when a take is bad rather than three days later


• Owning the metrics — throughput, yield, rework rate, inter‑annotator agreement, cost per usable episode — instrumented, visible, and trending the right way


• Running a continuous improvement loop with the team, not a one‑time setup

Data infrastructure at petabyte scale

• Storage, ingest, and processing for petabytes of episodic multimodal robot data, and the cost model that goes with it


• Indexing and retrieval so that any slice — task, skill, operator, site, failure mode, date — is one query away


• Dataset curation and versioning — reproducible training sets, filtering, dedup


• Build versus buy decisions across the stack, and long‑term ownership of both


• Lineage, retention, and access control across two entities and multiple sites

Qualifications

• Ph.D, M.S. or B.S. in Computer Science or a related field.


• 5+ years building data infrastructure or ML data platforms in production, including time spent owning a human‑in‑the‑loop data pipeline end to end.


• Preferred Experience: Standing up or scaling an in‑house or outsourced data collection or annotation operation — writing the process, building the tooling, training the people, and owning the quality numbers.


• Experience with petabyte‑scale multimodal data — video, sensor, and time‑series — including storage layout, chunked or columnar formats, and retrieval at training throughput.


• Robotics or AV data experience — ROS bags, MCAP, multi‑sensor time synchronization, calibration, episode‑level metadata.


• Minimum Strong Python plus one systems language, and comfort owning services, storage, and pipelines end to end.


• Production experience with object storage, distributed processing, and data lake and indexing patterns (S3, Ray, Spark, Beam, Parquet, Iceberg or equivalents).


• A track record of shipping internal tools people actually used — full stack enough to put a usable UI in front of a pipeline.


• Evidence you are genuinely good with people and process, not just tolerant of them — you have trained a team, written the SOP, and then improved both.


• Obsessive about data quality. You have caught the subtle problem that everyone else’s dashboard said was fine.


• Willingness to work on real hardware and across time zones with the collection team. Our data comes off robots, and you will spend time in the lab.

Benefits

• Competitive equity package with significant upside


• Comprehensive health, dental, and vision insurance


• Flexible PTO and remote work options


• Latest hardware and compute resources for experimentation

Company Culture

PhyAgents fosters an environment of ownership and visionary leadership. As a founding member, you will have the autonomy to shape the technical architecture and define the technical+research roadmap. The culture is fast-paced, emphasizing rigorous experimental design and real-world impact. We value engineers who are not just builders but also leaders who can set technical direction and mentor others. Diversity and inclusion are core to our mission, ensuring that we build robust models for diverse industrial environments.

Location

San Jose, California, USA

"