Applied Research Engineer – Video Data & ML

Role Overview

We are seeking an Applied Research Engineer with a strong foundation in video understanding, machine learning, or computer vision to help improve the quality of video datasets powering state-of-the-art AI models. This role is perfect for a candidate with 3–5 years of experience in ML/AI who is eager to deepen their skills through hands-on dataset development and small-model fine-tuning under the mentorship of senior engineers and researchers.

You’ll work with ML teams, QA leads, and delivery managers to design precise, benchmark-aligned video annotation pipelines, contribute to small-scale model experiments, and enhance labeling workflows that directly support real-world AI systems. Strong cross-functional communication will be key to translating modeling goals into actionable annotation strategies.

Key Responsibilities

ML-Aligned Data Development

• Co-develop clear, structured guidelines for video annotation tasks including:
  
  
• Frame-level and segment-level classification
  
  
• Temporal localization and gesture/action recognition
  
  
• Multi-object tracking across frames and scenes
  
  
• Human-object and multi-agent interaction labeling
  
  
• Work with ML stakeholders to align labeling specs with downstream use cases such as action classification, event detection, and object tracking.
  
  

Benchmark-Driven Data Optimization

• Identify labeling gaps affecting model performance on public benchmarks (e.g., MVBench, LongVideoBench, Video-MME, AVA-Bench).
  
  
• Recommend guideline updates based on error analysis and metric improvements.
  
  

Model Collaboration & Fine-Tuning

• Support small-scale model fine-tuning efforts (e.g., vision transformers or temporal CNNs) under the guidance of senior engineers.
  
  
• Run basic evaluation experiments to assess annotation quality and model impact.
  
  

QA and Labeling Process Support

• Collaborate with QA leads to build gold sets, spot-check protocols, and error rubrics that improve consistency and reduce ambiguity.
  
  
• Help close the loop on annotation feedback through structured escalation and review.
  
  

Cross-Functional Communication

• Act as a technical bridge between ML engineers, annotators, and QA reviewers.
  
  
• Create clear documentation and communicate updates across technical and non-technical stakeholders.
  
  

Qualifications

• 3–5 years of experience in computer vision, applied ML, or data-centric AI, especially involving video data or temporal modeling.
  
  
• Working knowledge of video modeling techniques and benchmarks for tasks like tracking, segmentation, or action recognition.
  
  
• Some hands-on experience with fine-tuning or evaluating small ML models using tools like PyTorch, TensorFlow, or Hugging Face.
  
  
• Familiarity with video labeling tools (e.g., CVAT, VOTT, Labelbox, SuperAnnotate) or experience working with custom platforms.
  
  
• Strong understanding of the ML data lifecycle, including synthetic data, annotation QA, and human-in-the-loop systems.
  
  
• Ability to interpret relevant research papers and translate them into actionable annotation or modeling improvements.
  
  
• Excellent communication skills—comfortable presenting ideas, gathering requirements, and collaborating across teams.
  
  

What Success Looks Like

• Scalable, reproducible annotation pipelines that improve model accuracy on benchmark tasks.
  
  
• Well-documented workflows that reduce ambiguity and inter-annotator error.
  
  
• Hands-on contributions to model fine-tuning and evaluation under mentorship.
  
  
• Constructive collaboration across ML, data, and QA teams that accelerates iteration and deployment readiness.