Release notes for Alibaba Cloud Large Language Model Engineer ACA Certification

Updated at:
Copy as MD

This topic describes the latest updates for the Alibaba Cloud Large Language Model Engineer ACA Certification.

2026

V2.4.2 Adds Graph Engineering content and optimizes prompt engineering and RAG courses

Date: July 2026

Graph orchestration: How are roles and work divided? How does data flow? Where is manual approval required? The industry refers to this design work as Graph Engineering. This update adds this content to the "Add a self-check loop to an agent" lesson. It covers key topics in graph orchestration, such as how to define nodes, how states transition, where to set manual checkpoints, and when not to use a graph.

Prompts: When you use a Large Language Model (LLM) to generate text, the response may seem complete but not entirely accurate. This issue often occurs because the goal, approach, and boundaries are not clearly defined. This update restructures the "Improve response quality by optimizing input" lesson. It adds workflows for segmented and collaborative execution of complex tasks, such as interviews, research, planning, and self-checks. This is in addition to the existing content on element completion and example alignment. The lesson also explains how to convert a successful solution into a reusable prompt template and validator to create a repeatable loop.

Key optimizations

  • Adds an introduction to concepts related to graph orchestration.

  • Adds an experiment: Consolidate the weekly report generation workflow into a Skill.

  • Restructures the prompt engineering course to include workflows for completing complex tasks using LLMs.

  • Optimizes the retrieval-augmented generation (RAG) course content to improve the introduction to topics such as vector retrieval.

New graph orchestration content

Key optimizations to the prompt engineering course

Optimized RAG course content

image

image

image

V2.4.1 Adds Loop Engineering content

Date: June 2026

The term Loop Engineering is now used in the industry to describe feedback-driven, automated agent loops. This update adds the term to the "Add a self-check loop to an agent" lesson. It explains how to transform the 'generate, check, revise, re-check' cycle from a manual, repetitive process into a controllable, automated workflow.

Automate the self-check loop

image

V2.4 Course structure upgrade: From "Understanding LLMs" to "Mastering AI Agents"

Date: May 2026

In early 2026, Coding Agents, such as Qwen Code, OpenClaw, and QoderWork, expanded from developer tools into general office applications. Mastering AI is becoming a core skill for all employees. This upgrade shifts the course focus from 'learning to use LLMs' to 'building a personal AI work system.' It focuses on how to control, manage, and collaborate with agents. The new version restructures all 11 lessons along an 'Understand → Communicate → Act → Master → Implement' learning path. This helps students grow from AI users into designers of AI work systems.

Core changes

Upgrade dimension

Content

Upgraded course focus

Shifts from explaining basic LLM applications to focusing on how to control, manage, and collaborate with agents. All 12 lessons are restructured accordingly.

Upgraded practical methods

Introduces real-world work scenarios, such as desktop agents, knowledge bases, tools, MCP, Skills, and self-check loops.

Upgraded implementation skills

Strengthens business objectives, evaluation mechanisms, and the RIDE methodology to help students turn AI capabilities into executable solutions.

Course structure changes

Lesson

After update

Change description

1

Understanding LLMs

Updates the AI development timeline, introduces the agent concept, and aligns with the Alibaba Cloud product ecosystem.

2

Typical application scenarios for LLMs

The lesson is restructured into a three-layer application model: generation assistant, execution agent, and decision support.

3

Improve response quality by optimizing input

Helps students master methods for high-quality communication with LLMs.

4

Enable LLMs to answer previously unanswerable questions

Explains how to build an enterprise knowledge base service and ensure service quality through business objectives and evaluation mechanisms.

5

Improve LLM performance in vertical domains through fine-tuning

Retains the fine-tuning chapter, focusing on applicable scenarios, basic processes, and the criteria for choosing between prompts, RAG, and fine-tuning.

6

Install a desktop agent and complete your first task

Shifts from a plug-in perspective to a desktop agent introduction, lowering the barrier to entry for non-technical students.

7

Connect agents to knowledge bases and tools

Explains how to use a personal AI agent for Q&A and introduces tool usage and MCP principles.

8

Solidify your work methods with Skills

Adds Agent-Skill content, explaining how to consolidate mature work methods into reusable Skills.

9

Add a self-check loop to an agent

Expands from agent task execution to include output verification, self-correction, and closed-loop improvement.

10

Security and compliance for LLM applications

Focuses on security risks in agent scenarios and adds five key principles.

11

Learning summary

The original extended learning content is split and merged into relevant lessons. The summary now connects with the enterprise AI implementation methodology.


V2.3 Updates the enterprise AI transformation methodology

Date: March 2026

In response to student feedback about uncertainty on how to implement the skills they have learned, this update adds an enterprise AI transformation practice framework to the learning summary. It introduces a methodological loop from 'scenario identification' to 'implementation validation.' This helps students transform the tools and methods from the course into executable AI application plans.

Updated content

Enterprise AI transformation practice framework

image.png


V2.2: RAG evaluation optimization and improvements

Date: January 2026

The original RAG section of the course focused on the setup process but lacked details on performance evaluation. This update adds basic methods for RAG performance evaluation and systematic improvement strategies. It guides students to build a complete understanding of the 'build → validate → optimize' cycle. This enhances their ability to independently diagnose and improve Q&A systems.

Updated content

Deep involvement of business experts in RAG evaluation

Root cause analysis approach for Q&A systems

image.png

image.png


2025

V2.1.1 Optimizes the fine-tuning chapter content

Date: December 2025

The original course's comparison of RAG and fine-tuning understated the value of RAG. This could mislead students into prioritizing fine-tuning. This update rewrites the comparison paragraphs. It clarifies that fine-tuning is more suitable for specific scenarios, such as when a deployment environment cannot provide retrieval services or when you want the model to internalize knowledge. This helps students make more accurate technology choices.


V2.1 Optimizes content across multiple lessons

Date: May 2025

To help students better understand key concepts, such as LLM working principles, tool calling, RAG, agents, and fine-tuning, this update optimizes the explanation order, diagrams, and case study presentations in several lessons. This improves the accuracy, coherence, and readability of the course content.

Key optimizations

  • Optimizes diagrams related to basic LLM concepts, the training process, and the product ecosystem.

  • Adjusts case studies and explanations in the chapters on prompts, RAG, tool calling, and agents.

  • Adds explanations for key content, such as RAG evaluation, agent structure, and fine-tuning methods.

Three stages of LLM training (Lesson 1)

image.png


V2.0 updated exam syllabus

Date: February 28, 2025

The original exam syllabus featured an imbalanced distribution of knowledge points and lacked practice questions. This update adjusts the weighting of exam topics and releases a practice exam to help students prepare more effectively. It also corrects readability and formatting issues in some course content.

Updated content

  • Optimizes the weighting of exam topics.

  • Releases practice exam questions.

  • Optimizes the readability and formatting of some content.

V1.0 Exam topics

Ratio

V2.0 Exam topics

Ratio

LLM concepts

12%

LLM concepts

18%

Prompt engineering

14%

Prompt engineering

20%

Enhancing LLM application capabilities

24%

Enhancing LLM application capabilities

20%

LLM fine-tuning

12%

LLM fine-tuning

12%

LLM Agent

20%

LLM Agent

14%

LLM security and compliance

12%

LLM security and compliance

10%

Multimodal LLMs

6%

Multimodal LLMs

6%


V1.5 Updates Lesson 1 and optimizes Lesson 4

Date: January 26, 2025

The Qwen model's capabilities are continuously updated, and the automated task case studies in the original course were not closely related to daily office scenarios. To help students better understand the model's capabilities and master reusable automated analysis methods, this update adjusts Lesson 1, 'Understanding LLMs,' and Lesson 4, 'Automate tasks with APIs.' The update to Lesson 1 revises the Qwen model introduction and adds scenario-based case studies. The update to Lesson 4 adds content on using Lingma for data analytics.

Qwen scenario-based case studies

Lingma data analytics

image.png

image.png


2024

V1.0 Course launch

Date: June 6, 2024

The first version of the Alibaba Cloud Large Language Model Associate Engineer (ACA) certification course was launched. The course is designed for individuals with non-technical backgrounds and beginners to LLMs. It is structured linearly by technical module and consists of 11 lessons. The lessons cover LLM concepts, application scenarios, API calls, prompts, plug-ins, RAG, fine-tuning, agents, security and compliance, multimodal extensions, and a learning summary.

The practice environment is based on the Alibaba Cloud Model Studio platform. The exam is an online test that consists of 50 questions: 35 single-choice and 15 multiple-choice. A score of 80 or higher is required to pass.