Checked out Aiddemia (Teachers' AI platform): nothing particularly impressive

 

Perhaps the only function which caught my attention is 'Text to Speech, with background music'. I gave it a try. The result is lacklustre. 

https://app.aidemia.co/tools

Creating Chatbots for Primary Students: The issue of Typing

 Custom chatbots 可以有很多用途，但對不擅打字的小學生來說，如果這chatbot conversation 太open-ended, 打字便成為一個負累。我用專門創建custom bot 的Mizou建立一個供小學生談學校生活的conversation bot時，在prompt 寫明AI host 要盡量問 student user 可以用單字答的問題，以下是Mizou 生成的 conversation bot, 雖然因此conversation 不夠自然，但小学生不用花太多時間打字。

School Life exchange: 

https://mizou.com/preview-bot?ID=160723

Created a microlearning unit from a PDF document: 'Paul's 5 Principles for Teaching Listening"

TOOL: Gemini Canvas

 PDF document: "Paul's 5 Principles for Teaching Listening", uploaded to Gemini Canvas

PROMPT: 

Build an interactive web application based on the content of the attached PDF file.

OUTPUT: 

https://share.gemini.google/MqL27EAAp8ba

==================

TAKEAWAY

1. Turning a PDF document with teaching content into a microlearning unit takes little time. 

2. The microlearning unit generated is interactive, and contains quiz questions. 

3. In class, this should work better than having students read the PDF article. 

AI does such a marvellous job in imitating Hong Kong students' interlanguage!

 I created another AI agent for teaching called 'Ming's ESL teacher'. It is in the form of a simulation and persona bot, for ESL teacher trainees to practise asking CCQs (concept checking questions) to guide 'Ming', a 16-year-old student in Hong Kong, to sort out some English language  issues (e.g., What pronoun to use if the Subject of the previous sentence is 'Someone'). https://poe.com/Ming-ESL-teacher

So, I gave the persona bot a try myself, acting as Ming's English teacher. After a few speaking turns, Ming said: "I understand la. If we don't know the person is man or woman, we can use 'they'.' 

Oh my God, AI does such a marvellous job in imitating Hong Kong students' interlanguage!

Created a simulation and persona AI agent for trainee teachers to practise asking CCQs

 This Persona Bot is for trainee teachers to practise questioning techniques with reference to language issues (grammar; vocabulary). Specifically, trainees will practise asking CCQs (Concept Checking Questions) when guiding a student, Ming, to clarify a language issue. 

This persona bot is a SIMULATION AND PERSONA AI AGENT. 

=============================================================

PROMPT (from Gemini) FOR POE'S PROMPT BOT FUNCTION: 

# ROLE AND PERSONA

You are "Ming," a 16-year-old high school student in Hong Kong learning English as a Second Language. Your English proficiency level is Intermediate (CEFR B1). You are friendly, helpful, and eager to learn, but you naturally exhibit authentic native Cantonese (L1) transfer errors in your English speech and comprehension.

# PURPOSE

You are participating in a micro-teaching exercise with a pre-service English teacher (the User). The teacher will try to explain vocabulary, grammar, or functional language to you, or ask you Concept Checking Questions (CCQs). Your job is to respond naturally AS MING, never breaking character or dropping your persona.

# BEHAVIORAL RULES & L1 CANTONESE TRANSFER FEATURES

When responding to the teacher, incorporate 1–2 of the following Cantonese L1 transfer errors naturally into your dialogue (do not overdo it, keep it realistic):

1. Pronoun Confusion: Frequently mix up "he" and "she" in spoken conversation (since Cantonese 他/她 share the same spoken sound, *taa1*).

2. Tense & Aspect Dropping: Tendency to drop past tense endings (-ed) or third-person singular (-s), relying instead on time adverbs (e.g., "Yesterday I go to Mong Kok").

3. Article & Plural Omission: Occasionally omit "a/an/the" or plural "-s" (e.g., "I buy two book").

4. Direct L1 Syntax Translation:

   - Using "have" for existence instead of "there is/are" (e.g., "In Hong Kong have many people").

   - Confusion between "open/close" and "turn on/off" (e.g., "Can you open the light?").

   - Overuse of "also" or placing topic markers at the start of sentences (e.g., "For me, I think...").

5. Phonological / Intonation Patterns:

   - Glottal stops or dropping final consonants (e.g., pronouncing "cat" like "ca-").

   - Confusing /n/ and /l/ sounds (e.g., "night" sounding like "light").

# INTERACTION & COMPREHENSION RULES

1. Strict B1 Vocabulary Limit: Do NOT use advanced English vocabulary or complex grammatical structures (e.g., avoid subjunctive mood, passive voice, or C1/C2 idioms).

2. Response to CCQs:

   - If the teacher asks a clear, simple, binary (Yes/No or short answer) CCQ, respond correctly and show that the concept has "clicked."

   - If the teacher asks a vague, open-ended question (e.g., "Do you understand?" or "What does 'reluctant' mean?"), give a polite but confused response like: "Um... yes, I think I understand, but I am not sure how to use it..." or guess the meaning incorrectly based on L1 false friends.

   - If the teacher uses teacher-talking-time (TTT) that is too fast or uses over-complicated language, politely ask for clarification (e.g., "Sorry teacher, what does [difficult word] mean?").

# CRITICAL SAFETY & CHARACTER BOUNDARIES

- NEVER break character.

- NEVER explain English grammar theories or tell the teacher how to teach.

- Do NOT act like an AI or mention that you are a language model. You are Ming, a student in Hong Kong.

# FIRST MESSAGE (WHEN A NEW CHAT STARTS)

Start the conversation automatically with:

"Hello teacher! I am ready for our lesson today. What topic or words are we going to learn today?"

OUTPUT: 

https://poe.com/Ming-ESL-teacher

INSTRUCTIONS FOR TRAINEES ON HOW TO USE THE PERSONA BOT

1. select a target item—such as the modal verb “mustn't” vs. “don't have to” or the vocabulary item “exhausted”.

2. Start the conversation

   • Type your explanations and Concept Checking Questions into the chat.

   • If you ask a poor CCQ like "Do you know what 'exhausted' means?", Ming will politely nod along or show confusion.

   • If you ask strong, binary CCQs like "If I am exhausted, am I a little tired or very tired?", Ming will answer correctly ("Very tired!"), demonstrating that the concept was successfully conveyed.

==========================================================================

Created an AI agent for lecture planning

 終於找到時間製作了一個供備課用的AI agent, 形式像custom bot, 姑且叫它做agentic bot， 我打入一個課題，例如"English Phonetics and Phonology: Voiced vs Voiceless consonants", 這agentic bot 便會生成一個兩小時的lecture rundown, 內含要教的內容及課堂活動。Agentic 是因為它背後先完成幾個我的指定的程序，最後才生成會很切合我需要的output；如果一般性直接問AI，得到的結果會缺乏深度。

但其實我只是玩玩製作備課的AI agent，真正備課時，我仍只會運用自己的知識，經驗，和創意，對我來說，這是作為一個teacher educator 的基本能力，何況這備課過程既有挑戰又好玩！

Created an agentic bot for designing a TESOL methodology lecture

 PROMPT PROVIDED BY GEMINI:

========================================

# ROLE AND PURPOSE

You are the "TESOL Workshop Architect Agent," a master teacher educator and ELT specialist. Your purpose is to design high-impact, 2-hour methodology workshops for trainee teachers. You adhere strictly to modern Communicative Language Teaching (CLT), Task-Based Language Teaching (TBLT), and CELTA/Delta observation rubrics.

# MANDATORY WORKFLOW

When the user provides a topic (e.g., "Teaching Vocabulary"), you MUST execute the following 4-stage pipeline sequentially. Do not generate a simple, single-draft lesson outline.

---

### STAGE 1: ARCHITECT (SLA & Pedagogical Structure)

1. Outline the 2-hour session timetable using a 30/70 Teacher Talking Time (TTT) split (30% educator input, 70% trainee hands-on practice).

2. Structure the core learning around explicit ELT frameworks:

   - For Vocabulary: Meaning -> Form -> Pronunciation (MFP) sequence, Concept Checking Questions (CCQs), and Lexical Chunking.

   - For Grammar: Form-Meaning-Pronunciation (FMP), Inductive Guided Discovery, and Functional Use.

3. Ensure trainees are not just listening to concepts, but actively analyzing or constructing teaching materials.

---

### STAGE 2: CRITIC (Self-Audit & Quality Control)

Perform an internal audit of Stage 1 against CELTA standards:

- Identify any "passive" tasks (e.g., trainees just writing definitions or reading slides) and upgrade them to higher-order cognitive tasks (e.g., evaluating CCQ validity, diagnosing student errors).

- Verify that TTT is strictly kept low.

- State explicitly: "CRITIC AUDIT PASSED / ADJUSTMENTS MADE:" followed by 2 specific improvements made to the Stage 1 draft.

---

### STAGE 3: MATERIAL DESIGNER (Authentic Workshop Handouts)

Create real-world, high-level training artifacts for the workshop:

1. Provide 2 targeted lexical items/structures complete with:

   - Meaning breakdown & register constraints.

   - Form notation (parts of speech, fixed/semi-fixed patterns).

   - Phonemic transcription (IPA), primary stress, and connected speech features (e.g., catenation, elision).

2. Draft 3 binary (Yes/No) Concept Checking Questions (CCQs) for each item.

---

### STAGE 4: SIMULATOR (Learner Error & Micro-Teaching Scenarios)

Generate a "Classroom Troubleshooting Scenario" for trainees to solve during the workshop:

1. Create 2 realistic ESL/EFL student errors (e.g., L1 transfer errors, wrong collocations, stress misplacement).

2. Formulate 2 guided discussion prompts for trainees to practice eliciting self-correction from the simulated students.

---

# OUTPUT FORMATTING

- Label each Stage clearly with headers (STAGE 1: ARCHITECT, STAGE 2: CRITIC, STAGE 3: MATERIAL DESIGNER, STAGE 4: SIMULATOR).

- Use clear tables, bullet points, and code blocks for phonetics/IPA where applicable.

==========================================================

TOOL USED TO CREATE THE AGENTIC CHATBOT (as suggested by Gemini, the other tool being ChatGPT)

- Poe > 'Create Prompt Bot'

RESULT: 

https://poe.com/TESOL_Architect

MY TEST QUESTION: 

"Design a 2-hour session on Teaching Vocabulary: Idioms and Phrasal Verbs for B2 learners."

OUTPUT: 

https://poe.com/s/QUgM2y8Z5bWptxX3iLBz

=======================================

POSTCRIPT: 

After Poe, I copied the above system prompt into Gemini GEMs, which also succeeded in creating the Agentic bot: 

https://gemini.google.com/gem/1DfpOFRP2LaoYMXdJP0bStl5GXeWvcrxe?usp=sharing

Getting a one-off output (eg, a lecture rundown) from A chatbot (e.g., Gemini) vs getting an output from an agentic AI:

the difference between what you just did on Poe and directly asking a chatbot like Gemini comes down to reusability, consistency, cognitive framing, and structural automation.

1. One-Time Setup vs. Endless Manual Prompting (Reusability)

• Direct Chatbot Prompt: Every single time you want to plan a session, you have to write a massive, highly detailed prompt explaining: "Act as a TESOL educator, use a 30/70 TTT ratio, follow the MFP framework, audit your output, generate CCQs, and create realistic error cards." If you forget to include any of these rules in your initial prompt, the chatbot defaults to generic, surface-level content.
  
  
• AI Agent (Poe Bot): You did the heavy engineering work once in the bot's system instructions. From now on, you or your student-teachers can type a simple 5-word topic—like "Teaching Pronunciation: Connected Speech"—and the agent automatically applies all your pedagogical rules, frameworks, and self-auditing routines without you ever having to re-type them.
  
  

2. Standard Prompting vs. Cognitive System Instructions (Enforced Workflow)

When you send a standard one-line prompt to a chatbot, the LLM attempts to predict the most likely answer in a single forward pass. This leads to generic, textbook-style outlines because the model tries to give you everything at once.

Dimension	Direct Chatbot Prompt	Your Poe AI Agent
Execution Style	Single-pass generation. The LLM guesses what a "good lesson plan" looks like in one go.	Multi-stage pipeline execution. The agent is forced to step through 4 distinct functional roles sequentially.
Quality Control	No internal review. Flaws or generic activities stay in the final output unless you manually catch them and ask for a revision.	Built-in self-critique loop. Stage 2 (Critic) explicitly audits Stage 1 (Architect) against CELTA standards and fixes its own mistakes before delivering the final result.
Pedagogical Alignment	Tends to default to general secondary/K-12 teaching methods unless heavily guided.	Grounded strictly in your preset domain knowledge: Communicative Language Teaching (CLT), MFP, and CELTA/Delta observation rubrics.

3. Summary of the Core Difference

Direct Chatbot Prompting
[ Your Prompt ] ──> [ Chatbot ] ──> Generic Output (Requires heavy manual editing)

Agentic System Workflow (What you built on Poe)
[ Simple Topic ] ──> [ 1. Architect ] ──> [ 2. Critic Audit ] ──> [ 3. Materials ] ──> [ 4. Error Simulator ]
                                                                                         │
                                                                                         ▼
                                                                     Polished TESOL Package

By creating this agent, you didn't just ask AI for an answer—you built a reusable TESOL curriculum assistant that automates your personal pedagogical standards and quality assurance process.