ALLMAcademy

Curriculum

The Learning Path

Start with Fundamentals, then move through the 5 core stages. Open capability or production paths only when your current problem needs them. Prerequisites guide the order without turning the course into a maze.

15
Modules
32
Topics

Start Here

Build the shared mental model for model calls, context windows, retrieval, state, memory, tools, and agents.

4 lessons
Primer

Agent Engineering Fundamentals

Learn the system map before choosing prompts, retrieval, memory, tools, loops, graphs, or multiple agents.

4 lessonsStart here

Core Path at a Glance

Each stage contains the earlier ones
  1. 01Prompt Engineering
  2. 02Conversation Engineering
  3. 03Response Engineering
  4. 04Loop Engineering
  5. 05Graph Engineering

Core Stages

The shortest path from one model request to durable, graph-controlled systems.

5 modules
S1

Prompt Engineering

Design one clear model request with instructions, examples, constraints, and a realistic success test.

2 lessonsAfter Primer
S2

Conversation Engineering

Manage turns, roles, history, summaries, retrieval, and what carries forward between messages.

2 lessonsAfter S1
S3

Response Engineering

Turn stochastic text into typed, validated, repairable responses and safe tool requests.

2 lessonsAfter S2
S4

Loop Engineering

Build bounded model–tool loops with routing, retries, verification, budgets, and explicit exits.

4 lessonsAfter S3
S5

Graph Engineering

Make control flow explicit with typed state, branches, parallel nodes, cycles, subgraphs, and human gates.

2 lessonsAfter S4

Agent Capabilities

Reusable capabilities, interoperability, state, memory, and executable specifications.

3 modules
C1

Skills, Plugins & MCP

Distinguish capabilities from packaging, then connect hosts and servers through MCP.

2 lessonsAfter S3
C2

State, Memory & Durability

Separate stateless requests, thread state, checkpoints, durable tasks, and long-term memory.

2 lessonsAfter S4
C3

Spec-Driven Agents

Define goals, capabilities, invariants, states, exits, and verification before implementation.

2 lessonsAfter S4

Production Deep Dives

Serving, retrieval, evaluation, safety, economics, and strategy when you need more depth.

6 modules
D1

Inference & Serving

KV cache, prefill vs. decode, PagedAttention, continuous batching, and caching.

2 lessonsAfter S1
D2

Model Efficiency

Speculative decoding, quantization, and distillation through one bandwidth-and-quality lens.

3 lessonsAfter D1
D3

Retrieval & RAG

Semantic and lexical retrieval, hybrid search, GraphRAG, reranking, freshness, grounding, and evaluation.

4 lessonsAfter S2
D4

Evaluation & Observability

Regression, adversarial, judge, and human evaluations connected to traces and production drift.

2 lessonsAfter S3
D5

Production Safety & Cost

Cost attribution, prompt injection, authorization, isolation, and cache safety.

3 lessonsAfter S4
D6

Strategy & Tradeoffs

Choose adaptation, orchestration, and serving approaches across latency, quality, cost, and reliability.

1 lessonsAfter S5 + D4