Forward Deployed

Staff AI System Design: Building an Enterprise AI Platform: sections and modules

T18

Staff AI System Design: Building an Enterprise AI Platform

Grow one enterprise AI platform from an MVP to a multi-region, staff-level architecture, one forced change at a time.

Requirement: Optional
For whom: Senior and staff candidates who finished T17 and want depth
Before this track: AI System Design; T17 first
By: Sourav Banerjee

Objectives

Defend the whole platform in about a minute without notes
Draw the final architecture from memory, with cell route authority, signed generations and the control-plane outage path
Write an ADR that removes options breaking a hard constraint before ranking, and compute a migration break-even by hand
Diagnose a rejected cutover or a lagging cell: name the cause, the safe action and the owner

The smallest system that works, then work that survives failure.

T18.1
Enterprise use case and target architecture
5 lessons
5 lessons
T18.2
Build the MVP
5 lessons
5 lessons
T18.3
Distributed execution
5 lessons
5 lessons

Exercise: Stage labs: the in-browser walkthrough and experiment, and the Python lab
Done when: Every stage exit gate passed

Trusted retrieval, then growth and overload without unfairness.

T18.4
Enterprise knowledge
5 lessons
5 lessons
T18.5
Scale the application
5 lessons
5 lessons
T18.6
Handle overload and fair scheduling
5 lessons
5 lessons

Exercise: Stage labs: the in-browser walkthrough and experiment, and the Python lab
Done when: Every stage exit gate passed

Isolation, consistency and releases you can roll back.

T18.7
Enterprise security
5 lessons
5 lessons
T18.8
Reliability and consistency
5 lessons
5 lessons
T18.9
Production AI quality and releases
5 lessons
5 lessons

Exercise: Stage labs: the in-browser walkthrough and experiment, and the Python lab
Done when: Every stage exit gate passed

When to route, self-host or fine-tune, and how to govern it.

T18.10
Inference and GPU infrastructure
5 lessons
5 lessons
T18.11
Fine-tuning and model customisation
5 lessons
5 lessons

Exercise: Stage labs: the in-browser walkthrough and experiment, and the Python lab
Done when: Every stage exit gate passed

Regions, residency and the final architecture defence.

T18.12
Global multi-region platform
5 lessons
5 lessons
T18.13
Staff-level platform architecture
6 lessons
6 lessons

Exercise: Stage labs: the in-browser walkthrough and experiment, and the Python lab
Done when: Every stage exit gate passed

Every decision the stages take, with its context, options, consequences and reversal criteria, in one place.

T18.14
Architecture decision records
13 lessons
13 lessons

Exercise: Rewrite one ADR for a changed constraint and rescore it
Done when: You can defend each decision and name what would reverse it