PUBLIC ROADMAP

AI Systems Engineer roadmap

A metadata-only view of the sequence, milestones, and free-preview boundary.

AVAILABLE CURRICULUM: 64 units
Free preview: 3 free-preview units

Applied Programming and Backend Foundations for AI Systems

Foundations

Build the first working layer of AI Operations Copilot from a reproducible project environment through deterministic triage, validation, files, packaging, typing, tests, CLI, FastAPI, Pydantic, provider abstraction, and a controlled structured model output.

AI Operations CopilotSmart Inbox TriagePython + AI Backend Foundations

14 units

01
Environment, terminal, Git, and the first running component
A01 · Starter path
02
Python data, collections, and inbound-message representation
A02 · Starter path
03
Conditionals, loops, and the deterministic triage baseline
A03 · Starter path
04
Functions, responsibilities, and the triage pipeline
A04
Milestone
05
Validation, exceptions, and controlled failure
A05
06
TXT/JSON input and safe result persistence
A06
07
Package structure, imports, and pyproject.toml
A07
08
Type hints, enums, and explicit domain models
A08
09
Pytest and regression protection
A09
Milestone
10
CLI boundary: single and batch triage
A10
11
HTTP and the first FastAPI service
A11
12
Pydantic and POST /triage: transport → domain → response
A12
Milestone
13
Configuration, secrets, and a replaceable model provider
A13
14
Structured model output and the first hybrid AI triage demo
A14
Milestone

Your phase progress will appear here after account access is connected.

Persistent AI Backend and Operations Memory

Persistence

Turn the stateless Smart Inbox Triage service from Phase A into a durable backend with SQL, PostgreSQL, SQLAlchemy, transactions, repositories, migrations, persisted analysis runs, idempotent API behavior, filtering, pagination, indexes, and database integration tests.

AI Operations CopilotPersistent Operations MemorySQL + Persistent AI Backend

11 units

01
Relational thinking and SQL fundamentals
B15
02
Production schema design, constraints, joins, and transaction behavior
B16
03
SQLAlchemy 2.x engine and mapped persistence models
B17
Milestone
04
Sessions, commits, rollbacks, and transaction boundaries
B18
05
Repositories and persistence-independent services
B19
06
Alembic migrations and explicit schema evolution
B20
Milestone
07
Durable message ingestion and analysis-run lifecycle
B21
08
Idempotent write APIs and conflict behavior
B22
Milestone
09
History queries, filtering, sorting, and pagination
B23
10
Indexes, PostgreSQL integration tests, and database failure cases
B24
11
Persistent Operations Memory end-to-end
B25
Milestone

Your phase progress will appear here after account access is connected.

AI Evaluation, Reliability, and Release Decisions

Evaluation

Turn Persistent Operations Memory into an AI Quality Lab: define evaluation contracts, review a curated benchmark, freeze a holdout split, build a local evaluation harness, calculate classification and human-review metrics, analyze slices and failure modes, measure selective automation and execution reliability, compare one controlled candidate, test robustness, and make a release decision without moving the goalposts.

AI Operations CopilotAI Quality LabAI Evaluation + Reliability

10 units

01
Evaluation contracts: define what good means before measuring it
C26
02
Reviewed evaluation cases and label discipline
C27
03
Dev and frozen holdout splits with dataset hashing
C28
Milestone
04
Local evaluation harness, subjects, graders, and raw artifacts
C29
05
Confusion matrices, precision, recall, F1, and review safety metrics
C30
Milestone
06
Selective automation, execution reliability, latency, and cost
C31
07
Slice analysis and a useful failure taxonomy
C32
08
Robustness with metamorphic cases and repeated-run stability
C33
09
One controlled candidate experiment and a predeclared promotion gate
C34
Milestone
10
Frozen holdout decision and the AI Quality Lab capability
C35
Milestone

Your phase progress will appear here after account access is connected.

AI Workflows, Human Review, and Reliable Actions

Workflows

Turn the evaluated AI Operations Copilot into an operational workflow system with explicit state transitions, durable workflow history, human-review queues, approvals and overrides, optimistic concurrency, deterministic action planning, replaceable action adapters, idempotent HTTP execution, bounded retries, reviewer-feedback candidates, and end-to-end workflow metrics.

AI Operations CopilotOperations Workflow and Human ReviewWorkflows + Human Review

11 units

01
Workflow cases and an explicit state-machine contract
D36
02
Persist workflow cases, review decisions, events, and action requests
D37
03
Transition service and append-only workflow history
D38
Milestone
04
Build the human-review queue and operator-facing query API
D39
05
Human approvals, overrides, escalation, and review history
D40
Milestone
06
Optimistic concurrency and stale-review protection
D41
07
Deterministic action planning and the command boundary
D42
08
Replaceable action adapters and a controlled HTTP integration
D43
09
Idempotent action execution, bounded retries, and terminal failure
D44
Milestone
10
Turn human overrides into reviewed feedback candidates
D45
11
Operations Workflow end-to-end and workflow quality summary
D46
Milestone

Your phase progress will appear here after account access is connected.

Grounded Knowledge, RAG, and AI Security

Grounded Knowledge

Extend AI Operations Copilot with a safe and measurable knowledge layer: versioned document ingestion, deterministic chunking, PostgreSQL full-text retrieval, embeddings with pgvector, hybrid ranking, grounded answer drafts, citations, abstention, prompt-injection defenses, privacy controls, bounded failure handling, and separate retrieval, answer, security, and reliability release gates.

AI Operations CopilotGrounded Company IntelligenceRAG + Knowledge + AI Security

9 units

01
Knowledge contract, answerability, and threat boundaries
E47
02
Safe knowledge ingestion, provenance, and deterministic chunking
E48
Milestone
03
PostgreSQL lexical retrieval and a frozen retrieval benchmark
E49
04
Embeddings, pgvector, and hybrid retrieval
E50
Milestone
05
Grounded answer drafts, citations, and abstention
E51
06
Prompt injection, privacy, and secure context construction
E52
Milestone
07
Failure budgets, bounded retries, stale sources, and partial artifacts
E53
08
RAG evaluation, security gates, and retriever release decisions
E54
09
Grounded Company Intelligence end-to-end
E55
Milestone

Your phase progress will appear here after account access is connected.

Observability, Integration, and Deployment

Production

Turn AI Operations Copilot into a reproducible deployed portfolio system with release acceptance, privacy-safe structured logs, OpenTelemetry traces and metrics, health and readiness checks, Docker, PostgreSQL-backed integration tests, GitHub Actions CI, a thin operator control room, a restricted public demo surface, deployment smoke tests, and a final technical release gate.

AI Operations CopilotAI Operations Control RoomProduction + Observability

9 units

01
Full-system release contract and offline acceptance
F56
02
Privacy-safe structured logging and correlation IDs
F57
Milestone
03
OpenTelemetry traces and application metrics
F58
04
Health, readiness, operational summaries, and failure visibility
F59
05
Docker, Compose, and production-like PostgreSQL integration
F60
Milestone
06
GitHub Actions CI and migration-aware quality gates
F61
07
AI Operations Control Room and restricted demo surface
F62
Milestone
08
Container deployment, managed PostgreSQL, and smoke verification
F63
09
Final technical release, reproducibility, and system verification
F64
Milestone

Your phase progress will appear here after account access is connected.

AI Systems Engineer Find your path

Loading workspace

Applied Programming and Backend Foundations for AI Systems

Persistent AI Backend and Operations Memory

AI Evaluation, Reliability, and Release Decisions

AI Workflows, Human Review, and Reliable Actions

Grounded Knowledge, RAG, and AI Security

Observability, Integration, and Deployment