Title: Surgical Video Data & Annotation Operations Manager
Location: Pune
About Us
At Codvo, we are committed to building scalable, future-ready data platforms that power business impact. We believe in a culture of innovation, collaboration, and growth, where engineers can experiment, learn, and thrive. Join us to be part of a team that solves complex data challenges with creativity and cutting-edge technology.
Role Summary
Own and scale a production pipeline that ingests, de-identifies, annotates, and delivers 10,000+ hours/month of surgical video with ≥70% automation and audit-ready compliance. You will design the ontology, stand up model-in-the-loop labeling, run QA/IAA, and hit SLAs from ingest → labeled → approved datasets for training and evaluation.
Core Responsibilities
1) Program Ownership & Throughput
• Deliver ≥10,000 labeled video hours/month within SLA (≤7 days ingest→approved).
• Maintain ≥70% auto-accept rate via model-assisted labeling; drive to 80%+ over time.
• Operate daily pipeline: ingest → de-id → pre-labels → annotation → QA → release.
• Manage backlog, staffing, and shift planning to sustain 333+ hours/day throughput.
2) Ontology & Guidelines
• Define and version surgical ontologies: phases, steps, events, tools, anatomy, quality flags.
• Author annotation guidelines with boundary rules and ambiguity handling.
• Run change control and backward compatibility across dataset versions.
3) Tooling & Automation
• Stand up and operate CVAT (or equivalent) with API-driven workflows.
• Integrate MONAI Label (or similar) for model-assisted segmentation/active learning.
• Use FiftyOne (or equivalent) for dataset QA, error analysis, and sampling.
• Implement propagation (tracking/interpolation) and confidence-based routing.
• Partner with ML to deploy pre-label models (tool detection, phase recognition, SAM-style masks).
4) Quality, IAA & Release Gates
• Define gold sets and acceptance thresholds:
• Phase/event F1 ≥ 0.92
• Tool mAP ≥ 0.85
• Anatomy Dice ≥ 0.85
• Run double-labeling (≥10%), adjudication, and weekly QA reviews.