Teachers rely on Microsoft Teams and OneNote to collect student writing, reflections, and project work, yet extracting that work for grading, feedback, or research is often burdensome. This presentation demonstrates a privacy-conscious “Diary Pipeline” that uses the Microsoft Graph API to export OneNote pages into a CSV/XLSX schema with consistent headers and rich metadata, including page and section, student ID, timestamps, photo URLs and counts, diary text content and statistics, feedback fields, vocabulary entries.

Attendees will see how a single class notebook can be exported quickly, then reused for multiple purposes: faster grading and rubric scoring, portfolio content, longitudinal tracking, and formative feedback or LLM-assisted review. Student diaries are used as a model for varied forms of data export. The focus is on non-invasive data gathering that lightens the burden on both instructors and students.

This poster presentation will explain how to create text analysers using Claude AI to write the code necessary. With some specific prompt engineering, the AI can be trained to produce tools that give highly specific feedback in formats that are clean and crisp. With only a few simple steps in the vibe-coding process, the poster will demonstrate both early (simple scoring and word counting) and more developed examples (descriptive feedback comments, APA corrections, Fry Readability and CEFR scores) that were created in just a few minutes to great effect (detailed tailored feedback output). Attendees who bring an internet-ready device should be able to create a running analyzer on the spot and test it out immediately. The presentation will also show examples of output from each of the engineered analyzers with clearly defined limitations and possibilities of the tool Claude AI that is free to use and easily adaptable and embeddable to your service of choice or within its own system. No previous coding skills are necessary, but the created code will be displayed for all to see and share via QR codes.

Providing feedback on student writing is time-intensive, but using online generative-AI tools like ChatGPT raises concerns about student privacy and teacher accountability. This session reports on an open-source, offline feedback assistant that runs on teachers’ desktops/laptops using small, local large language models (LLMs). The goal is to reduce time spent on writing feedback without putting student work online while maintaining pedagogically appropriate feedback for different proficiency levels. Methodologically, the development of the assistant required a purpose-built benchmark to compare candidate LLMs on one-shot judgement tasks, such as judging the effectiveness of topic sentences and aligning claims and evidence accurately. Results from the one-shot tests on controlled, synthetic learner texts show that tiny models like the one-billion parameter TinyLlama model frequently hallucinate and make logical errors. Even larger local models, such as the 20-billion parameter GPT-OSS-20B model, prefer overly academic registers in their feedback, reducing suitability for lower-proficiency learners. For CALL practice, the findings highlight a key trade-off: privacy-focused offline feedback demands nuanced GenAI engineering but cannot substitute for teacher oversight. Teachers who wish to preserve student privacy will need new skills to review, correct and contextualize model output and to have an explicit understanding of model limitations.

With the prevalence of social media and short attention spans, young people are increasingly consuming information in bite-sized pieces, rather than engaging with longer texts, leading to shallower reading. Specifically designed and built by language experts to counter these issues and featuring over 1,700 animated stories aligned to CEFR, BOOKR integrates voice recognition, recommended reading lists and AI-generated assessment tools to promote student engagement, assess comprehension and provide a personalized learning experience at scale.

This session introduces MiMi Chat, a Gen AI-powered tutor that serves as a conversation partner and provider of formative feedback and assessment, tightly integrated with classroom-based English language curricula. Now used at over 50 universities worldwide, MiMi gives students structured opportunities to practice speaking and receive real-time feedback aligned with CEFR “CAN-DO” goals—outside scheduled class time but directly connected to in-class instruction. Drawing on data from over 15,000 students, we examine measurable gains in speaking output and learner confidence. Case studies include use in a TED-based discussion course, a presentation course, a nursing communication module, and a cross-cultural communication program. We also share engagement metrics, feedback accuracy, and qualitative learner insights.

This hands-on workshop introduces an AI-assisted app development method centered on real-time voice interaction with large language models. Participants will observe and then practice a structured voice-first workflow for early-stage development: articulating constraints and success criteria aloud, iteratively refining requirements, and having the model generate and explain small, usable code components for classroom-ready CALL tools. Voice is used intentionally for ideation and specification, while code generation and implementation steps, such as testing, debugging, and deployment, are demonstrated using standard on-screen workflows. Successful examples, such as a randomized speaking-partner scheduler or an automated PDF region-extraction tool, draw on educator-built systems in Japanese secondary and tertiary settings, with attention to constraints faced by non-specialist programmers and common institutional limitations. Attendees will prototype a simple classroom tool or research utility with the explicit goal of building something they can “use on Monday.” Aligned with the 2026 theme “Prevail or Fail?”, the workshop emphasizes practical judgment: when voice interaction accelerates design and when it introduces friction. Participants leave with a working prototype, a workflow checklist, and reusable prompt templates. Laptop with Wi-Fi required; a smartphone or personal hotspot may be helpful for connectivity redundancy. A phone/headset is recommended.

AI should assist and enhance human-centered language teaching and learning, not replace it. Established principles in second language acquisition remain central: learners need sustained exposure to comprehensible input, meaningful opportunities for output and interaction, structured fluency development, and timely feedback. Eigo.AI is designed around these principles. The platform offers a comprehensive library of AI-generated, human-proofread lessons across proficiency levels. Each lesson integrates listening, reading, speaking, and writing through seven structured activity types, ensuring balanced skill development. Students receive immediate feedback on pronunciation, writing, and discussion performance, while teachers maintain full oversight through detailed tracking tools.

Eigo.AI also addresses a practical constraint in most programs: time. Reaching advanced proficiency typically requires more than 2,500 hours of focused study, far beyond what classroom instruction alone can provide. The platform extends structured learning beyond class hours while keeping progress visible and measurable. Teachers can monitor engagement, review student output, and intervene when needed. In this way, AI supports informed instructional decisions without displacing teacher expertise. This session will demonstrate how Eigo.AI combines established pedagogy with practical implementation, offering institutions a scalable and classroom-ready solution that strengthens learner development and preserves teacher agency.

We are developing an AI-powered, web-based system to analyze the nonverbal communication (NVC) skills of EFL students in Japan. Teaching verbal communication (VC) skills is challenging due to typical student-to-teacher ratios of 40:1, with NVC requiring even more individualized instruction and feedback. In this case study, we focus on the AI-enhanced video analysis tool we are developing to support the final system. We analyze the facial expressions (perceived emotions and gaze/engagement) and body language (head and facial movements) of EFL students (aged 18–20) using standard webcams and correlate these data with online assessments from a diverse group of human judges. We conduct mock (simulated) online job interviews, as this method allows us to achieve two goals: 1) primarily collect data to improve the functionality of our NVC training system, and 2) secondarily provide students with feedback that can assist them in other online interview contexts, such as TOEFL, IELTS, and future academic or professional interviews. We are developing our final system in-house and will showcase our software methodology and its current developmental stage. Once complete, our system will enhance students' NVC skills by increasing their awareness, motivation, and self-efficacy, while improving their online interview performance.

Generative AI offers new opportunities for second-language absolute beginners, but maintaining linguistic control over AI-generated output remains a challenge. This presentation introduces and evaluates an AI-based tool designed to address this issue.

The tool was introduced in a Spanish entry-level course after three weeks of instruction and is continuously updated to align with curricular progression. To date, it has provided on-demand reading comprehension activities to more than 700 learners at Xi’an Jiaotong-Liverpool University (China). The system operates as an AI agent that integrates multiple workflows and does not require model training or fine-tuning. Instead, it builds on the authors’ previous research on prompting techniques for enforcing linguistic constraints.

The presentation focuses on two areas. First, it outlines the tool’s design strategies for maintaining linguistic control, managing complexity, and increasing output variability. Second, it reports findings from an evaluation of 360 AI-generated texts produced across different curriculum stages and text types.

Results show high levels of lexical and grammatical control, providing evidence that linguistic control can be scaffolded and sustained in AI tools for absolute beginners. Attendees will gain practical insights into designing AI systems for absolute beginner-level instruction without programming backgrounds, with applications transferable to other languages and proficiency levels.