Skip to content
Linear Horizon Training

Ready-made course · 2 days · Advanced

AI Platform Engineering

A two-day course for platform engineers, architects and senior developers: build the internal infrastructure that lets multiple product teams use AI safely and consistently.

If every product team integrates a model, a vector store and a logging story independently, the organisation gets duplicated cost, inconsistent security and no way to evaluate quality. Platform engineering for AI is the work of making a paved path: model access, policy, retrieval, observability and developer experience.

This course covers model gateways, routing, authentication, policy, rate limiting and model abstraction. Prompt management, evaluation, secrets, data controls, RAG services, vector infrastructure, cost allocation, caching, resilience and developer experience are designed as platform products.

The teaching is design-heavy, with enough implementation detail to be concrete. You work through the contracts a platform should offer a product team, and the ones it should refuse to hide.

You leave with a service design, not a catalogue of tools to install.

Who it is for
Platform engineers, architects and senior developers responsible for shared AI infrastructure and developer enablement.
Before you start
Experience building internal platforms, APIs or shared infrastructure. Familiarity with LLM applications is expected. This is not a first course in generative AI.

You will be able to

  • Design a model gateway with authentication, routing, policy and rate limiting.
  • Specify model abstraction and prompt management without hiding evaluation from product teams.
  • Provide RAG and vector infrastructure as a service with tenancy and data controls.
  • Implement observability, cost allocation and caching as platform features.
  • Define secrets, resilience and fallback behaviour that product teams can rely on.
  • Shape developer experience: SDKs, golden paths, documentation and a review process.
  • Draw a responsibility boundary between platform and product that can be operated.

Modules

8 quarter-day modules. Each one can go into a custom course on its own.

  1. The AI gateway and control plane

    • One point of access to every model provider.
    • Keys, quotas, routing and fallbacks.
    • Logging, redaction and policy at the gateway.
    • Build or buy: open-source and managed gateways.
  2. Model catalogue and access policy

    • An approved model catalogue, and how models get onto it.
    • Access by team, data class and environment.
    • Deprecations and migrations across many consumers.
    • Exercise: write the access policy for your organisation.
  3. Retrieval as a platform service

    • Shared ingestion, indexes and connectors.
    • Permissions enforced by the platform, not each app.
    • Self-service for teams with guardrails.
    • Running cost and who pays it.
  4. Policy, data controls and audit

    • Redaction, classification and policy enforcement.
    • Central logging and retention.
    • Evidence for auditors and regulators.
    • Policy as code.
  5. Evaluation as a platform service

    • Shared evaluation tooling and datasets.
    • Release gates teams can use without ceremony.
    • Comparing models for many teams at once.
    • Production monitoring as a service.
  6. Developer experience for AI platforms

    • Templates, SDKs and golden paths.
    • Local development against platform services.
    • Documentation and support that scale.
    • Measuring time to a first production feature.
  7. Cost allocation and FinOps for AI

    • Tagging and attributing every model call.
    • Budgets, alerts and showback.
    • Commitments, provisioned capacity and discounts.
    • Cost per outcome, not cost per token.
  8. Running the AI platform

    • SLOs for the platform and for its providers.
    • Incidents: provider outages, bad releases and abuse.
    • Capacity, quotas and fairness between teams.
    • A platform roadmap driven by its users.

Practical work

Design exercises throughout. You specify gateway, retrieval, evaluation and cost-allocation services, critique a weak platform proposal, and produce a paved-path design including developer experience. Implementation snippets are used where they clarify an interface; the deliverable is a platform design.

You take away

  • A model-gateway and paved-path design
  • Service contracts for retrieval, evaluation and cost allocation
  • A platform versus product responsibility matrix
  • Developer-experience and SLO notes

Combine it with

Contact