Job description
About the Role
We're a small, fast-moving AI productivity startup (~25 people) building an autonomous AI executive assistant that operates across email, calendars, meetings, and business software. This role owns the measurement system that determines whether our AI agent is genuinely improving in ambiguous, real-world environments.
You'll partner closely with AI Agent Capabilities engineers to produce the evidence that drives product decisions, model choices, and release quality — turning hard questions about agent behavior into rigorous, actionable answers.
Visa sponsorship is not available for this role.
What You'll Do
Architect and maintain automated evaluation pipelines that measure agent quality across product surfaces.
Translate product capabilities into explicit pass, partial-pass, and failure criteria for complex multi-step tasks.
Build representative gold datasets and regression suites covering real workflows, edge cases, and adversarial scenarios.
Define and track metrics including task success, tool-selection accuracy, instruction adherence, factual consistency, latency, cost, and reliability.
Design deterministic and model-based graders, calibrate LLM-as-a-judge systems, and monitor grader agreement.
Compare models, prompts, and implementations using rigorous offline experiments and production evidence.
Analyze traces and production outcomes to identify root causes and build a practical failure taxonomy.
Convert production failures into regression cases and continuously close gaps in evaluation coverage.
Build dashboards and release-quality signals that make results actionable for engineering, product, and leadership.
Partner with capability engineers to recommend improvements and verify that fixes raise quality without introducing unacceptable regressions.
What We're Looking For
Required experience (dealbreakers):
4+ years in Applied Data Science or Machine Learning roles, with a focus on building and delivering evaluation systems, automated data pipelines, or production ML infrastructure.
Demonstrated experience designing and implementing automated evaluation frameworks, success criteria, and regression suites for complex AI/ML or agentic systems.
Production-grade proficiency in Python and SQL, with hands-on experience building and maintaining automated analytical pipelines on large datasets.
Core requirements:
Experience applying statistical and experimental methods — significance testing, variance analysis, sampling — to evaluate non-deterministic AI/ML systems.
Experience developing labeled datasets, annotation guidelines, and quality-control processes for ground-truth data in dynamic product environments.
Deep understanding of LLM agent behaviors including tool use, multi-step execution, retrieval, and practical failure modes.
Ability to analyze model traces, tool calls, and outputs to identify root causes of failures across model, prompt, tool, and data layers.
Experience using production telemetry and observability data to monitor system quality, build dashboards, and analyze real-world user outcomes.
Nice to have:
Prior hands-on experience with LLM-as-a-judge systems, model-based grading, or AI benchmarking platforms.
Experience shipping or operating production ML products, agentic systems, or customer-facing consumer software.
Experience reviewing and adapting public research benchmarks or academic evaluation methodologies to real-world product problems.
Key traits we value: strong product orientation (prioritizing metrics tied to real user outcomes over convenient proxies), high ownership, analytical rigor, and comfort driving ambiguous quality questions from design through to product decisions.
Location
This role is based in Palo Alto, CA. Visa sponsorship is not available.
Originally posted on Himalayas
Who can apply
Eligible countries: United States. Accepted UTC offsets: UTC-10, UTC-9, UTC-8, UTC-7, UTC-6, UTC-5, UTC+14. Review the full description for employer-specific work authorization, residency and schedule requirements.