Senior Product Operations Manager, Evaluation Quality

Harvey•San Francisco, CA

4d•$155,400 - $233,200

About The Position

We’re looking for a senior operator to own the quality bar behind Harvey’s human evaluations. As we scale globally, the volume of eval work is growing 10x, but volume only matters if the output is trusted. This role makes Human Data’s signal decision-grade: rigorous, calibrated, and reproducible enough that Product, Engineering, and AI Research act on it to ship. As a member of our Evaluation Operations team, you’ll work alongside our Evaluation Operations Manager (who runs throughput and coordination) and partner closely with Applied Legal Researchers, Product, Engineering, and AI Research. You'll set the standard for what "good" looks like across eval data and methodology, and certify that results are sound before teams rely on them, building the stakeholder trust that lets EPD act on the signal.

Requirements

6+ years in product operations, research operations, evaluation/QA operations, or quality program management
A track record of owning quality inputs (guidelines, instructions, benchmarks, QA procedures) for complex, expert-driven or human-in-the-loop work
Experience onboarding, training, and calibrating a distributed pool of expert raters, annotators, or reviewers, and running the feedback loop that improves their quality over time
Enough grounding in measurement concepts (calibration, inter-rater reliability, sampling, rubric design) to independently set up and own quality yourself
Comfort with interpreting evaluation data, natively or with AI tool support
A record of scaling and streamlining quality processes under shipping pressure, with a bias toward documentation and reproducibility over one-off analysis
Ability to work deeply with domain experts (e.g., ALR / lawyers) and translate nuanced judgment into repeatable, documented standards
Strong cross-functional coordination across Product, Engineering, Research, ALR, and data providers/vendors
Clear communicator who can build credibility and trust with stakeholders
Bias to action and high ownership, from writing the guideline to auditing a vendor batch line by line

Nice To Haves

Experience in legal tech or working with domain experts in regulated industries
Experience owning quality across multiple markets, languages, or jurisdictions
Built calibration, inter-rater reliability, or capability-tracking systems for annotation or evaluation pipelines
Experience transitioning evaluation work in-house or otherwise improving evaluation ROI
Familiarity with LLM-as-judge / automated evaluation used alongside human eval
Early employee at a hyper-growth startup, or experience at a world-class product or platform operations org

Responsibilities

Own the quality bar for Harvey’s human evaluations: define what “good” looks like for eval data and make outputs decision-grade for EPD
Author and maintain the evaluation guidelines, instructions, benchmarks, and gold references that contract attorneys work from, and serve as the arbiter on ambiguous or disputed judgments
Standardize and streamline rubric and evaluation design into repeatable templates and one documented methodology, partnering with Applied Legal Research (ALR), who supplies feature-specific legal depth
Own contract-attorney quality: onboarding, calibration training, inter-rater reliability, and the feedback loop that keeps judgment consistent across attorneys and over time
Define and maintain the failure-mode / error taxonomy that turns raw eval results into a structured, prioritized signal Product and ALR can act on
Run QA on vendor and contract-attorney deliverables against a defined bar before results inform a launch decision
Ensure the quality bar holds across jurisdictions and non-English geographies as coverage expands
Establish one standard, documented way to analyze eval results, and build lightweight operational dashboards to track rater capability and eval-program health
Support ALR in a review step that certifies an evaluation is sound before it scales to contract attorneys
Partner with ALR and Analytics to determine where human eval aligns with online signal and where it can provide expanded insights