Destava

Research Engineer - Data Eval

Calder is hiring for Research Engineer - Data Eval in Redwood City, CA, USA. Mid level. First seen November 7, 2026; open as of November 19, 2026. Jobs taken straight from company career pages — no aggregator in between.

Save to applyFree forever. No commitment, no card required.

Company
Calder
Where
Redwood City, CA, USA
Type
Full-time
Level
Mid
First seen
November 7, 2026
Listed via
Ashby
Open for
11 days, as of November 19, 2026

About this job

This position is accountable for quantifying data quality for embodied AI training datasets, employing ablation studies and model-driven verification on real-world data. It is part of the research group at a seed-stage company, encompassing the creation of assessment infrastructure and staff development. The role is located in Redwood City, CA.

Responsibilities

  • Execute ablation studies via training, fine-tuning, and evaluating open-source world models and robotic policies to assess the impact of collection, curation, and labeling decisions on final performance and against public references.
  • Employ models to inspect raw data for identifying poor-quality, corrupted, or atypical captures, detecting problems such as blur, exposure errors, missing frames, sensor misalignment, calibration shifts, and SLAM or stereo-depth failures promptly.
  • Contribute to shaping the evaluation research strategy, pinpointing critical unresolved questions for embodied and physical AI and conducting rigorous experiments.
  • Create benchmarks, metrics, and approaches that examine how leading robotic and world models perceive, reason, and execute actions.
  • Manage model evaluation comprehensively, covering offline analyses, A/B and regression tests, and real-world deployment across varied sensor configurations, rare edge scenarios, and practical tasks in manipulation, navigation, and interaction.
  • Develop pipelines, dashboards, and tools to ensure evaluation is efficient, reproducible, and mostly automated for both internal teams and external collaborators.
  • Search for edge cases, failure modes, and safety, privacy, and risk concerns, and relay insights to research and robotics groups.
  • Offer direction and mentorship across interdisciplinary initiatives while expanding the evaluation unit.
  • Act swiftly and assume responsibility, converting recent publications into prototypes and tools that operate at scale on actual sensor and trajectory data.
  • Guide technically, mentor organically, and collaborate effectively across research, robotics, engineering, and product domains.
  • Articulate clearly, document comprehensively, and excel in the uncertainty and speed of an early-stage environment.

Requirements

  • Track record of delivering impactful research in dynamic contexts, demonstrated by first-author publications at forums like NeurIPS, ICML, ICLR, CVPR, CoRL, RSS, ICRA, or IROS.
  • Strong mathematical grounding and substantial expertise in at least one domain adjacent to physical AI assessment, including reward modeling for embodied agents (covering reward hacking and specification gaming), calibration and uncertainty estimation for perception and policy models, simulation-to-
  • Practical familiarity with contemporary ML systems in both research and production settings, with notable experience assessing physical AI like robotic manipulation, navigation, world models, vision-language-action models, or multimodal egocentric perception.
  • Proficient in Python with robust engineering capabilities, adept at constructing scalable data pipelines, evaluation frameworks for simulated and real-world settings, and metrics that clarify embodied model quality.
  • Doctorate in Computer Science, Machine Learning, Robotics, Computer Vision, or a related field, preferably with a thesis or research emphasis on evaluation-adjacent subjects such as reward modeling, preference learning, human feedback, calibration, or automated assessment for embodied or perception
  • At ease with the ML experimentation stack (PyTorch, training and evaluation infrastructure, open-source model ecosystems), meticulous about experimental design, and inherently critical.

Nice to have

  • Familiarity with robotics, imitation learning, world models, computer vision models, or large-scale data pipelines is a significant advantage.

Read more and apply on Calder's site

Questions about this job

When was this Research Engineer - Data Eval job at Calder posted?

Destava first saw it on November 7, 2026, on Calder's own careers page (Ashby).

Is Calder still accepting applications for it?

It was open on Calder's own site when Destava last read it, on November 19, 2026, 11 days after we first saw it. When the company takes it down, this page says the job has closed.

How do I apply for Research Engineer - Data Eval at Calder?

On Calder's own site, at jobs.ashbyhq.com: open the posting. Destava does not take the application. Save the job to your Destava board to track it, see how well you fit and tailor your résumé to it.

Is this Calder job posting legit?

Destava read it on Calder's own careers page at jobs.ashbyhq.com, run on Ashby, not on a job board. First seen November 7, 2026. Apply on the company's own site, and never pay to apply for a job.

What other jobs is Calder hiring for?

Calder has 13 open postings on its own site: see all jobs at Calder.

How many computer and information research scientists jobs are open in San Francisco Bay Area?

768 open, 82 of them new this week, at 286 companies: see them all.

Applying to this one?

See how well you fit, before you apply

Save it to your board. Add your résumé once, and Destava shows where you are strong, what is missing and how to pitch yourself, then tailors your résumé to this posting.

  1. Found it
  2. Save it
  3. See your fit

Free forever. No commitment, no card required.

Or apply on Calder's own site

With Destava: Save to apply · Tailor my résumé · Write a cover letter

This page includes information from the O*NET 31.0 Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA). Used under the CC BY 4.0 license. O*NET® is a trademark of USDOL/ETA.