Xrosters

The Problem

Employee rostering seems simple until you realise that every shift interacts with labour laws. Award rates, overtime thresholds, minimum rest periods, casual vs. part-time vs. full-time rules, penalty rates for weekends and public holidays, break entitlements — all of these compose into thousands of rule combinations that determine whether a roster is legal and what each employee gets paid.

The manager needed to see potential breaches and pay consequences while assigning shifts, before the roster reached payroll.

What We Built

Xrosters evaluates every shift assignment against the configured pay and compliance rules before it is committed. If a shift would breach a labour law rule, the system warns the manager. If it would trigger a higher pay rate, it shows the cost impact in real time.

Key Modules

1. Roster Builder — drag-and-drop scheduling with live compliance checking. The system flags potential breaches as you assign shifts.
2. Labour Law Compliance Engine — a rules engine that models award rates, overtime, penalties, minimum breaks, and rest periods. Rules compose and conflict — the engine resolves them.
3. Facial Recognition Attendance — employees clock in and out via face scan. The system computes actual hours worked, not scheduled hours.
4. Wage Computation — combines rostered shifts, actual attendance, and the applicable configured rules to compute wages automatically.
5. Employee Self-Service — casual and part-time employees can submit availability, swap shifts with colleagues, give away shifts when sick, and request leave.
6. Onboarding/Offboarding — digital employee lifecycle management with document storage and compliance tracking.
7. Notifications — the system notifies managers when an employee's work hours might breach a labour law rule or trigger minimum wage adjustments.

Architecture

The compliance engine is central to Xrosters. It's a rule-based system where:

• Rules are composable (a single shift can trigger multiple rules simultaneously)
• Conflicts are detected and resolved (this version selected the most beneficial outcome when two configured rules conflicted)
• Rules are configurable per award, per state, per employment type
• Evaluation is real-time — the UI updates as you build the roster

The facial recognition module uses TensorFlow for face detection and embedding extraction. Face templates are stored as vectors, not images, for privacy.

What We Learned

The difficult part was translating pay conditions into rules that could be evaluated together. A shift can trigger several rules, so their interactions need to be modelled and tested as carefully as each rule on its own. The engine depends on a correctly maintained rule set; its checks are not a guarantee of legal compliance.

→ Engineering deep dive: Modelling 10,000+ rule combinations for automated wage calculation