Role Overview

• Responsible for planning and executing MiroMind’s data production and annotation projects (text / image / audio / video / multimodal). This includes organizing and managing internal and external annotation teams and vendors, establishing standards and processes, and ensuring the balance of progress, quality, cost, and compliance.

Key Responsibilities

Project Planning & Delivery

• Break down data requirements based on model training goals (task types, scale, coverage, difficulty, and priority). Develop milestones, budgets, and resource plans.

• Manage schedules, risks, and dependencies across concurrent data projects to ensure on-time delivery aligned with training needs.

Team & Vendor Management

• Build and mentor annotation and quality inspection teams (full-time / part-time / crowdsourced / vendors), including scheduling, performance, and incentive management.

• Manage vendor onboarding and evaluation (bidding, SLA, pricing, delivery quality), as well as cost and contract management.

Standards & Processes (SOPs)

• Define and iterate annotation guidelines, label taxonomies, edge cases, and decision trees; maintain operation manuals and case libraries.

• Design layered quality control (self-check, peer review, expert sampling), gold standard sets, and re-review workflows to continuously reduce rework rates.

Quality & Data Governance

• Establish quality metrics: gold standard accuracy, IAA (e.g., Cohen’s kappa / Krippendorff’s alpha), coverage, noise rate, PII leakage rate, etc.

• Apply methods such as active learning, hard example mining, weak supervision, and LLM-as-judge to drive a “data flywheel” for continuous refinement and augmentation.