Skip to main content

Challenge instructions

Instruction and application
In Progress

By the end of the hackathon, each team will deliver a 5-minute presentation, followed by 3 minutes of questions, covering:

  • The deployed ML solution, its architecture, and where it runs.
  • The monitoring plan and drift-detection strategy that keeps it healthy.
  • The lifecycle strategy including retraining triggers and decommissioning criteria.
  • Key risks, explainability findings, and stakeholder-tailored recommendations.

Success criteria​

  • A working deploy artefact — a live URL, a runnable docker-compose harness, or a reproducible container image.
  • A justified deployment architecture, including version control and a rollback strategy.
  • Evidence of secure, cost-aware, and appropriately scalable resource choices.
  • Explainability tooling (SHAP or LIME) wired in against at least one representative prediction.
  • Clear reproduction steps captured in a README that a stranger could follow.

Today, you will deploy your Phase 2 model into a production or test environment, version it, and produce a reproducible artefact that Workshop 2 will monitor.

Hackathon Resource Pack​

If you haven't already, or if you need to re-download it, download the Hackathon Resource Pack now. It contains:

  • Descriptions of the two scenarios.
  • Reference model artefacts for each scenario.
  • Deployment harness — docker-compose starter and hosted-URL deployment template.
  • Monitoring plan template for Workshop 2.
  • The stakeholder presentation template you will use throughout the hackathon.

You'll continue using this pack across both workshops 1 and 2.

AIMLF Hackathon 3 Resource Pack.zip

Once downloaded, open the files and keep them accessible for the rest of the hackathon workshop.

Working deployment reminder​

Use the deployment harness for your chosen scenario to guide your packaging, deployment, versioning, and explainability activities throughout workshops 1 and 2.

Save key outputs, configuration files, and reproduction steps from the harness to support your final presentation and handover documentation.

### Instructions

Package the model​

As a team:

  • Load your refined Phase 2 model into the deployment harness provided.
  • Pin dependencies and produce a reproducible container image or environment.

You may use LLMs or Atlas to:

  • Clarify container and packaging concepts.
  • Review Dockerfile / docker-compose choices.
  • Explore hosting options that fit the scenario's cost and traffic profile.

Deploy the artefact​

Deploy the model into a production or test environment. Choose one of:

  • A live URL exposing an inference endpoint.

  • A runnable docker-compose harness peers can start locally.

  • A reproducible container image and clear run instructions. Capture:

  • The deployment target and justification.

  • Version tags for the model artefact and the deployment configuration.

  • A short rollback procedure — how you would revert to a previous version.

Wire in explainability​

Add SHAP or LIME to the inference flow so that at least one representative prediction can be explained.

Focus on:

  • Showing the top features driving a single prediction.
  • Producing an artefact (chart, table, or JSON) you can include in the presentation.

You may use LLMs to:

  • Compare SHAP vs LIME trade-offs for your model type.
  • Suggest how to present feature attributions to non-technical audiences.
  • Help interpret unexpected feature importance results.

Capture deployment observations​

Use the presentation template to document:

  • Architecture and hosting choices, and why you chose them.
  • Security, scalability, and cost considerations.
  • Version control and rollback plan.
  • Early explainability findings.

Next steps in the hackathon​

In the next hackathon workshop, you'll add monitoring and drift detection, define your lifecycle and decommissioning plan, and prepare a stakeholder-tailored 5-minute presentation.

In progress