Strategic Project Lead – Software Engineering

What You'll Do10

• Run software-engineering data and evaluation programs from start to finish, including coding data, SWE benchmarks, and model evaluations.
• Source, calibrate, onboard, and manage contributors and reviewers, keeping reviewer capacity matched to program demand.
• Track throughput, quality, rework, defects, cost, and delivery timelines, then report on each.
• Inspect code, tests, task metadata, and final artifacts for technical correctness before release.
• Make release or rejection calls on submitted work and document the reasoning behind them.
• Diagnose quality failures and fix task design, rubrics, instructions, and review processes at the root.
• Lead recovery plans when a program falls behind on quality, throughput, cost, or schedule.
• Coordinate distributed software engineers, reviewers, and team leads across time zones.
• Report progress, risks, and corrective actions to customer research and program teams.
• Use Python, SQL, or similar tooling for quality analysis, defect tracking, and operational reporting.

Requirements7

• Direct experience running production software-engineering data programs for frontier AI models.
• Background in at least one of: coding training data, coding RLHF, SWE benchmarks, model evaluations, agentic coding, code generation, or code repair.
• Ownership of the full production system, from contributor sourcing and calibration through throughput, quality, rework, cost, and delivery recovery.
• Strong software-engineering judgment: read code and tests, inspect final artifacts, spot invalid technical work, and make release or rejection decisions with little ramp-up.
• Experience managing distributed software-engineering contributors, reviewers, and team leads while holding quality and delivery steady.
• Ability to take over a live program fast, diagnose operational problems without hand-holding, and explain decisions and recovery plans to demanding customers.
• Working knowledge of Python or SQL for analysis and reporting.

Who Should Apply

Candidates who have run coding data, RLHF, or SWE benchmark programs for AI labs and can still read code and tests themselves. The role suits someone who has taken over a struggling program and turned around quality, throughput, and cost. You need eight hours a day with four hours overlapping PST.