Production Manager, Lilt

• People management in AI data operations: 5+ years in AI/ML data operations or production, including 2+ years directly managing project managers or team leads in a distributed, multi-time-zone contractor environment
• LLM knowledge: Strong understanding of LLM training processes (pre-training, SFT, RLHF) and evaluation methodologies (human-in-the-loop, red teaming), and of what drives quality and throughput in annotation workflows
• KPI-driven management: Has run teams against throughput, quality (accuracy, IAA, gold-set), and cost-per-task targets; advanced proficiency with spreadsheets and dashboards, and able to use SQL to extract and analyze performance data
• Delivery track record: Has sustained on-time delivery and acceptance targets across multiple concurrent data collection or evaluation programs for enterprise or research lab customers
• Contractor workforce operations: Has hired, ramped, performance-managed, and offboarded hourly and freelance staff across regions and languages
• Methodology: Proven track record using Agile, Scrum, or Kanban to manage complex workflows across a portfolio of programs
• Communication: Writes clear, unambiguous guidelines and feedback for multilingual audiences and communicates status, risk, and tradeoffs crisply to leadership

Desirable

• PM Performance
• Outcome: Every active program has accountable Project Manager(s), and every PM carries a workload within the agreed span. Programs hit on-time delivery and first-pass acceptance targets without escalation; each PM is reviewed monthly against a scorecard of throughput, quality, and cost-per-task
• PM Hiring, Onboarding, and Development
• Outcome: PM pool capacity keeps pace with signed demand: new PMs are sourced, onboarded, and running their first program within the agreed ramp window, all programs start on time, and PM attrition is below threshold
• Quality Interventions Across Programs
• Outcome: Quality dips are caught mid-program through QA loops and corrected via retraining of annotator pools or guideline updates. Repeated misses by a PM or annotator pool lead to documented remediation or replacement. Issues are proactively discovered
• Process Standardization & Software Improvements
• Outcome: Programs launch from shared playbooks, guideline templates, and dashboard standards rather than being rebuilt per engagement. Every post-mortem produces documented improvements that lead directly into our custom software stack, and time from program handoff to first delivery declines quarter over quarter
• Escalation and Cross-Functional Interface
• Outcome: Risks surface to Technical Program Managers early enough to be managed. Escalations between the PM pool, Quality, Talent, and Delivery are resolved within agreed timelines