About the Company
Skyfall is disrupting the entire AI ecosystem by building the first world model for the enterprise. The goal of the ‘Enterprise World Model’ is to overcome the severe limitations of LLMs (Safety, Hallucinations, Expensive training) in order to provide enterprises significant value by having a comprehensive understanding of the complex interplay between data, people, and processes within organizations.
The Skyfall founding team consists of Maluuba founders who were previously pioneers in the Deep Learning revolution. Maluuba worked with AI pioneers such as Yoshua Bengio and Richard Sutton before it was acquired by Microsoft for $160M and became Microsoft’s AI research center in Canada.
About the Role
You will help build the training dataset that teaches our world model how humans use software. Our models learn to understand and predict computer-use workflows from real screen recordings, and the quality of that learning is entirely dependent on the quality of the data you curate.
Working alongside the world modeling team, you will review large volumes of computer-use videos, verify AI-proposed labels, separate genuine task workflows from non-workflow content, and flag noise and edge cases. This is precise, high-volume, judgment-driven work: every label you produce becomes ground truth the model trusts. A small, careless error can compound through training into a real failure in the deployed agent so accuracy, consistency, and integrity matter far more than raw speed.
Key Responsibilities
- Verify AI-proposed taxonomy labels (Domain, Subdomain, Software) on sampled videos, marking each as correct or incorrect against the labeling guide.
- Review video clips to determine whether they show genuine computer-use workflows (step-by-step task completion) versus non-workflow content (talking-head commentary, passive browsing, discussion), following defined labeling criteria.
- Annotate visual noise in sampled video frames (zoom, overlays, cursor effects, callouts, transitions, etc.), applying one or more labels per frame per guidelines.
- Work through shared, trackable annotation buckets across datasets ranging from 1,000–5,000+ videos.
- Flag ambiguous cases, edge cases, or tooling issues to the team rather than guessing, leaving clear notes to support escalation.
- Maintain consistent accuracy and pace against team-set quality and throughput targets, adapting quickly as guidelines evolve.
- Surface gaps or ambiguities in the labeling guidelines and taxonomy so they can be refined over time.
Core Requirements (Must-Haves)
Attention to Detail, Judgment & Integrity
- Exceptional attention to detail and the ability to stay consistent across long, repetitive, high-volume work.
- Ability to follow detailed, evolving labeling guidelines precisely and apply them uniformly.
- Sound judgment to distinguish genuine workflows from non-workflow content and to recognize edge cases.
- High integrity: prioritizes accuracy over throughput, and escalates ambiguity rather than guessing or forcing a label.
Software & Technical Fluency
- Strong general computer literacy and comfort navigating desktop operating systems (Windows, macOS, or Linux), web browsers, and common productivity, creative, and developer software, enough to intuitively recognize what a real task workflow looks like.
- Comfortable working with web-based annotation tools, spreadsheets, and trackers.
- Clear written English to document flags, notes, and edge cases unambiguously.
Qualifications
- Bachelor’s degree in any discipline (Computer Science, IT, or Engineering is a plus).
- 0–3 years in data annotation, QA / manual software testing, IT operations, or a similarly detail-oriented, guideline-driven role. Strong freshers with clear aptitude will be considered.
- Available to work full-time on-site at HSR Layout, Bengaluru for the duration of the 3-month contract.
Preferred (nice to have):
- Prior experience in AI data operations, video/content auditing, e-commerce/retail operations, or medical coding i.e., backgrounds that reward diligence under repetition.
- Basic understanding of how AI/ML models learn from training data.
- Familiarity with structured data formats (labels, JSON, taxonomies).
Who Will Succeed in This Role
- Someone who is meticulous, patient, process-driven, and reliable under repetitive, high-focus work.
- Someone who is just as precise and careful on their 400th video as on their 4th.
- Someone who treats ambiguity as a signal to pause and escalate, not an excuse to guess.
What This Role Is Not
- It is not a software engineering, development, or programming role.
- It is not a creative, content-production, or design role.
- It is not suited to those who only use computers casually; it requires real desktop-software fluency and comfort with high-volume, repetitive precision work.
Pay: Up to ₹50,000.00 per month
Education:
Work Location: In person
