Advanced Prompt Engineering & AI Evaluation is a course in designing prompts systematically for consistent results, covering chain-of-thought, few-shot examples, structured output and reducing hallucination, then measuring quality with golden sets, rubrics and LLM-as-a-judge. It suits regular AI users, developers, AI engineers and teams building AI workflows or agents.
Many organizations have started using generative AI in real work but hit the same problem: inconsistent output, different answers each time from the same request, quality that depends on whoever wrote the prompt, and no way to measure whether a revised prompt is actually better. These problems keep AI from scaling into standardized organizational processes, because both a systematic prompt-design technique and a scientific way to evaluate output quality are missing. This course raises prompt engineering from trial and error to measurable prompt engineering. Learners cover advanced prompt structure and patterns, reasoning techniques (Chain of Thought, ReAct, Tree of Thoughts), controlling
output into reusable formats (structured output), managing context and reducing hallucination, and finally the often-overlooked core of AI evaluation: building test datasets, measuring quality with rubrics and the LLM-as-a-judge technique, running regression tests on prompts, and setting up a shared, standardized prompt library. All content is taught through workshops on real tasks, covers ChatGPT, Claude and Gemini, and finishes with a capstone project where learners build their own end-to-end prompt system with an evaluation harness. (2 days, 6 hours per day, 12 hours in total, Intermediate to Advanced level, prior generative-AI use recommended, no programming required.)
Course Details
A 2-day course, 6 hours per day (12 hours in total), delivered as a hands-on workshop with a capstone project. Intermediate to Advanced level, covering ChatGPT, Claude and Gemini. Learners take home prompt patterns, a template library, rubrics and evaluation sheets, and a prompt security and PDPA checklist.
Section 1: LLM Foundations for Prompt Design
Section 2: Advanced Prompt Structure and Patterns
Section 3: Reasoning Techniques
Section 4: Few-shot and Example Engineering
Section 5: Controlling Output with Structured Output
Section 6: Context Engineering and Reducing Hallucination
Section 7: System Prompts, Templates and Prompt Library
Section 8: AI Evaluation Principles and Building a Test Dataset
Section 9: Evaluation Techniques and Regression Testing
Section 10: Prompt Security, Robustness and PDPA
Section 11: Capstone Project and Real-world Adoption
No public rounds are open right now. Join the waiting list and we will contact you first when the next round opens, or ask us on LINE. Or call 02-570-8449 or 088-807-9770
Chain-of-thought prompting asks the AI to reason step by step before giving its answer, which improves accuracy on problems involving several stages of logic or calculation. Few-shot prompting includes a handful of example inputs and desired outputs so the AI copies the format, tone and judgement criteria. They can be combined, and the examples should cover genuinely different cases rather than near-duplicates.
LLM-as-a-judge uses a second language model to score answers against set criteria. Write a clear rubric first, covering points such as accuracy, completeness and format, with a description of each score level. The judge then reads the question, the answer and the rubric and gives a score with its reasoning. Compare a sample against human ratings regularly to check the judge is not biased or drifting.
A golden set is a test dataset of real inputs with correct answers or criteria, covering normal cases, edge cases and cases that have failed before. Whenever you change a prompt, run every item in the set and score the results with a rubric or LLM-as-a-judge against the previous version. Adopt the new prompt only if the overall score improves and nothing that used to pass now fails, so you know the change genuinely helps.
It suits regular AI users who want consistent results, developers, data analysts and AI engineers putting LLMs into real systems, product owners designing AI features, and teams building chatbots or AI agents. You should have used ChatGPT, Claude or Gemini before. Programming is not required, although being able to read JSON or having called an API helps with structured output and evaluation.
You will be able to design prompts that give consistent results using advanced structures and patterns, apply chain-of-thought, ReAct and few-shot techniques appropriately, constrain output to a JSON schema, reduce hallucination through grounding, build golden sets and measure quality with rubrics and LLM-as-a-judge, run regression tests and defend against prompt injection. Participants who complete the training receive a Certificate of Completion with its own number, which can be verified online.
Yes. It can be run in-house for your team, with examples adapted to your own work, such as building test sets and quality criteria around the prompts or AI workflows your team already uses. You can request a quotation from this page, or contact us on LINE @itgenius or by phone on 02-570-8449.
Many organizations have already invested in AI tool licenses for their staff, only to find that actual usage is low. Most employees use AI only for general question-and-answer and never connect it to their daily work, so the return does not justify the…
As AI moves from a question-answering chatbot to an AI agent that actually does the work, Claude is one of the standout tools right now - especially for working with long documents, deep reasoning, and the new Cowork Mode that lets Claude access folders on…
Today many office workers do repetitive work across dozens of Excel files, wait months for the IT team to build small tools, or have good ideas they want to try but hit the wall of not knowing how to code. The arrival of AI coding agents like Claude Code…
