Skip to content
Ameen Abozaid
SRB-021

AI platform capabilities

  • 7 Lessons
  • 4 h practice (estimate)
On this page 9 topics

Overview

Topics

  • Shared LLM gateway with quotas
  • redaction
  • cost attribution
  • and routing
  • shared retrieval services
  • AI golden path

Learn first

4 h of practice (estimate)

Learning outcomes

  1. Design a shared LLM gateway: authentication, per-team quotas, redaction of sensitive data, logging, cost attribution, and model routing.
  2. Offer shared retrieval services (embedding, indexing, search) with clear data ownership per team.
  3. Build an AI golden path: template, evaluation harness, observability, and security checks pre-wired.
  4. Decide which AI capabilities to centralize and which to leave to product teams.

Lessons

  1. 01Why AI needs platform support
  2. 02The LLM gateway
  3. 03Quotas, redaction, and cost attribution
  4. 04Shared retrieval services and data ownership
  5. 05An AI golden path
  6. 06Demo: four teams, one gateway, one cost report
  7. 07Demo: redacting customer data before it reaches a model

Lab and exercises

Lab

SRB-021-LAB1 · FulfillCo LLM gateway

Outcomes: 1, 2, 3, 4 · Time: ~4 h

  1. Define gateway requirements from the four teams.
  2. Prototype auth, quotas, redaction, request logging, and cost attribution.
  3. Test a runaway team against another team's quota and scan logs for personal data.
  4. Write the AI golden path and decide what stays central versus product-owned.

Acceptance: one team's runaway usage cannot exhaust another's quota; cost per team reported; no unredacted personal data in logs.

Exercises

IDOutcomeTask
SRB-021-EX011List the gateway responsibilities and three non-goals.
SRB-021-EX021Design quota tiers for four teams with burst limits.
SRB-021-EX032Design a shared retrieval service with per-team data ownership and access control.
SRB-021-EX043Write redaction rules and five test inputs including tricky ones.
SRB-021-EX053Draft the AI golden path checklist.
SRB-021-EX064Decide which of six capabilities to centralize (gateway, prompt library, retrieval engine, eval harness, product prompts, UX).
Show solutions
  • EX01 Auth, quotas, redaction, logging, cost, routing; not prompts, evaluation of product quality, or business logic.
  • EX02 Per-team daily token budgets, per-minute rate limits, alerts at 80%, hard stop with clear error.
  • EX03 Namespaces per team, source permission filter at query time, deletion path, re-embedding.
  • EX04 Email, phone, address, order number patterns; tests for formats and false positives.
  • EX05 Gateway client, eval harness, observability, security checks, runbook.
  • EX06 Centralize gateway, retrieval machinery, eval harness; leave product prompts and UX to teams; prompt library optional.

Workplace evidence: a shared AI capability other teams adopted.

Quiz

Outcomes: O1 gateway · O2 retrieval services · O3 AI golden path · O4 centralize or not.

A self-check for you: nothing is graded, saved or sent anywhere. Choose an answer, then check it. You can try again as often as you like.

  1. 1A gateway primarily centralizes:O1
  2. 2Per-team provider keys cause:O1
  3. 3Quota isolation ensures:O1
  4. 4Cost attribution requires:O1
  5. 5In a shared index, data is owned by:O2
  6. 6Retrieval access control should:O2
  7. 7An AI golden path includes:O3
  8. 8Redaction must happen:O3
  9. 9Which should stay with product teams?O4
  10. 10Centralizing prompt design is:O4

Practical part

Platform capability review: present the gateway and golden path and defend central versus product-owned choices. Pass: 70% on the quiz and the three lab acceptance checks demonstrated.

Common mistakes

Every team holding its own provider keys; logging full prompts with customer data; centralizing prompt design that belongs to product teams.

Assessment

Platform capability review.

Workplace evidence

A shared AI capability other teams adopted.

Career relevance

AI strategy (SRA-021) and AI governance (STF-021).

Type to search levels, phases, modules and lessons.