Invigilo Technologies Pte. Ltd. – AI Operations Engineer

Role Summary

We are looking for an AI Operations Engineer to own the day-to-day reliability, quality, and scalability of our production AI systems. This role sits at the intersection of data, annotation, and production ML. You will manage training datasets, guide and QA annotation teams, investigate and resolve production issues caused by data or configuration, and build small tools and scripts to keep our AI pipelines running smoothly.

This is a hands-on, execution-focused role for someone technically competent (SQL, basic Python) who enjoys debugging real-world AI systems rather than only building models.

Key Responsibilities

• Own and manage datasets used for training and evaluation

• Curate, version, and document datasets

• Define data splits, sampling strategies, and quality checks

• Lead and support annotation operations

• Create and maintain annotation guidelines

• Train and instruct annotators on new classes and edge cases

• Perform QA and resolve ambiguous or incorrect labels

• Investigate and resolve production ML issues

• Debug drops in model accuracy or unexpected behaviour

• Identify root causes in data, configuration, or pipelines

• Work with engineers to deploy fixes and validate results

• Data analysis and troubleshooting

• Use SQL to query production databases and analyze incidents

• Detect data drift, class imbalance, and annotation gaps

• Automation and tooling

• Write small Python scripts and utilities to clean data, transform labels, and run batch analyses (you can leverage AI coding assistants)

• Build repeatable workflows for dataset preparation and evaluation

• Configuration and release support

• Manage model and pipeline configs across environments

• Validate new model releases against defined metrics before rollout

• Cross-team coordination

• Act as the bridge between ML engineers, software engineers, and annotation teams

• Communicate issues, priorities, and data requirements clearly

Required Skills

• Basic to intermediate SQL (joins, aggregations, filtering large datasets)

• Basic Python scripting (data processing, simple automation)

• Ability to use AI coding tools effectively to accelerate scripting and debugging

• Strong problem-solving and debugging mindset

• Good written documentation and communication skills

Nice to Have

• Experience with computer vision datasets and annotation

• Familiarity with ML/AI pipelines and model evaluation metrics

• Experience supporting production systems (logs, configs, monitoring)

• Exposure to tools like PostgreSQL, REST APIs, or message queues

What Success Looks Like

• High-quality, well-documented datasets that improve model performance

• Clear, consistent annotations with low rework and ambiguity

• Fast root-cause analysis and resolution of production ML issues

• Increasing automation that reduces manual data and labeling effort

This role is ideal for someone who enjoys being the operational backbone of an AI system: keeping data clean, labels accurate, and models behaving correctly in the real world as the company scales quickly.

Working Style & Hours

This is a startup environment with rapidly evolving priorities and frequent urgent deadlines.

Work is highly execution-focused and outcome-driven rather than strictly time-based.

Standard working hours are flexible, but you should expect periods of extended hours when investigating incidents, preparing urgent releases, or unblocking production issues.

Not a strict 9-to-6 role: responsiveness during critical production events is important.

You will often need to switch context quickly between debugging, data work, and coordinating with engineers and annotators.

Requires strong ownership mindset: when production quality drops, you take the lead in driving it back to green.

Comfortable with async communication and occasional off-hours coordination across teams when necessary to meet release timelines.

Emphasis on speed with discipline: ship fixes and improvements quickly while maintaining data and process quality.