Skip to content
Zester
All products

Occupational Health and Safety

SafetyFirst

An OH&S platform that predicts workplace incident probability over the next seven days, so prevention has somewhere specific to go.

Domain
Workplace Operations
Status
Deployed on IBM Watson
Stack
PythonXGBoostIBM WatsonFastAPI
Award
People's Choice Award, SaskHack 2026

7 days

Forward prediction window

The problem

Safety programs are measured on incidents that already happened.

Lagging indicators tell an organisation where it failed, months after the failure. The information that would have predicted the incident — who was on shift, what the conditions were, what the site's history looks like — exists at the time it would have been useful, and is never assembled into a probability anyone can act on.

How it works

  1. 01

    Assemble the real inputs

    Worker characteristics, current conditions, and historical patterns are brought together as one feature set rather than three separate reports.

  2. 02

    Model the probability

    An XGBoost model trained and deployed on IBM Watson scores incident probability over the following seven days.

  3. 03

    Serve it where work is planned

    Predictions are exposed through a FastAPI REST API, so the score reaches scheduling and briefing tools rather than a dashboard nobody opens.

  4. 04

    Act inside the window

    A seven-day horizon is chosen because it matches how shifts and site work are actually planned.

Architecture

  1. 01

    Inputs

    Worker characteristics, current site conditions, historical incident patterns

  2. 02

    Features

    Combined feature set aligned to a seven-day forward window

  3. 03

    Model

    XGBoost, trained and deployed on IBM Watson

  4. 04

    Serving

    FastAPI REST API returning incident probability

  5. 05

    Surface

    Predictions delivered into planning and briefing workflows

What changes

  • Safety attention is allocated by predicted risk rather than by last quarter's incident log.

    The model outputs a seven-day probability per context; the score is inspectable.

  • The prediction lands inside the planning window it is meant to change.

    Verifiable directly: predictions are served over a REST API at planning time.

  • The approach holds up to outside scrutiny.

    People's Choice Award, SaskHack 2026.

Capabilities

  • Seven-day incident probability prediction
  • Worker, condition, and history feature modelling
  • XGBoost model training and deployment
  • IBM Watson hosted inference
  • FastAPI REST prediction service
  • Integration into existing planning workflows

Who uses it

Safety officers

Deciding where to focus a limited number of interventions this week.

Site supervisors

Adjusting crew and task assignment against current conditions.

Operations leadership

Seeing risk concentration across sites before it becomes a claim.

Evidence pending

This page does not fabricate product imagery. The slots below are reserved at their final dimensions and name exactly what is needed.

Product screenshot required

Risk view for one site across the seven-day window.

Sample output required

A real prediction response from the REST API, redacted.

Have an operation this should be pointed at?