AI Operations Brain
Automated 90% of repeatable ops by turning incoming requests into approved AI skills and safe execution workflows.
MovieSaints · Internal Ops · AI Workflow Automation
At a glance
Built
Email-to-operations automation system for repeatable support and admin requests
Handled
Intent classification, skill lookup, approval gates, database updates, AWS/S3 context, and human escalation
Result
Automated 90% of repeatable ops while creating a supervised learning loop for new workflows
The problem
Most MovieSaints operational requests came through email: pricing changes, country availability, film metadata updates, support fixes, access corrections, reports, and similar tasks.
The issue was not that each request was technically hard. The issue was repetition, context switching, and risk. A small team had to read emails, interpret intent, look up the correct internal process, make database or system changes, and avoid breaking business rules.
The system needed to reduce manual operations without letting AI freely make unsafe production changes.
How it worked
Email intake
Requests sent to specific operational email addresses were ingested into the automation layer.
Intent classification
The AI determined whether the email was actionable, informational, ambiguous, or unsafe to automate.
Skill matching
The system checked whether an approved skill existed for the requested action, such as changing a film's price in a specific country.
Permission and confidence gates
Even when a skill existed, the system checked whether the action was approved for automation, whether inputs were complete, and whether confidence was high enough.
Safe execution
Approved workflows ran through defined internal tools, database rules, and infrastructure-aware scripts with traceable execution paths.
Human fallback
If the system could not find a matching skill, it routed the request to support and asked targeted questions about how the task should be performed.
Skill generation loop
After support explained a workflow, the AI inspected available system context such as database structure, AWS/S3 usage, and relevant code paths, then generated a new skill file for human approval when enough context was available.
Reuse
After approval, future similar requests could be handled automatically using the new skill and the same approval and audit controls.
Why it mattered
This moved operations from manual execution to supervised automation. It reduced repetitive operational work while preserving human control over new, ambiguous, or risky workflows.
The important part was not just automating individual tasks. The system created a way for repeated human instructions to become approved operational capabilities over time.
What this shows
This shows how I design AI systems that are useful in real operations: governed workflows with skill lookup, approval gates, execution paths, escalation, auditability, and learning loops.