AI Production Systems Core

Build, ship, operate, repair, and prove a bounded AI product capability.

12 calendar months · English

Outcome before content

Finish with a managed deployment, provider-portable source package, evaluation evidence, an injected incident repair, and a defended production decision.

Module 1

Model boundaries, tokens, and context

Diagnose token and context failure · Separate model capability from interface behavior · Select a bounded context strategy

Token-budget decision trace + Context-overflow repair evidence

Module 2

Instruction and system contracts

Define enforceable instruction boundaries · Separate trusted and untrusted instructions · Repair conflicting prompt contracts

Versioned system contract + Adversarial instruction test record

Module 3

Evidence, grounding, and abstention

Trace claims to evidence · Calibrate abstention and escalation · Diagnose unsupported answers

Claim-to-evidence ledger + Abstention escalation receipt

Module 4

Data quality, lineage, and leakage

Detect training and evaluation leakage · Maintain removal-aware lineage · Set data quality release gates

Lineage and removal record + Leakage-resistant split proof

Module 5

Model selection, serving cost, and latency

Compare model fitness under constraints · Calculate serving capacity and cost · Choose a measurable fallback

Model-selection decision record + Cost and latency envelope

Module 6

Evaluation and release evidence

Build a representative evaluation portfolio · Separate formative and release evidence · Detect evaluator failure

Versioned evaluation portfolio + Release evidence receipt

Module 7

Calibration, thresholds, and escalation

Calibrate confidence to consequences · Choose thresholds from asymmetric costs · Define human escalation boundaries

Threshold decision table + Calibration error analysis

Module 8

Authorization and retrieval security

Enforce subject and object authorization · Contain prompt and retrieval injection · Prove least-data access

Authorization test matrix + Retrieval attack and repair trace

Module 9

Observability, release, and incidents

Define actionable telemetry · Make evidence-backed release decisions · Contain and recover from incidents

Release or hold record + Incident containment and recovery runbook

Module 10

Core production mission

Integrate the Core controls · Deploy a portable production artifact · Repair an injected incident

Learner-owned deployment receipt + Independent incident repair evidence

Production validation is in progress

Public sales are disabled. No payment can be created until infrastructure, tax, legal, accessibility, and operational launch gates are independently approved.