For more than ten years, we’ve had the privilege of working closely with school districts and higher education institutions to help prepare the next generation of teachers, with a special focus on bilingual and ESL education. This collaboration has reinforced what we already knew: all educators need preparation tools that go beyond generic practice and truly mirror the challenges of teaching in today’s classrooms—whether that’s managing diverse learners, navigating real instructional scenarios, or succeeding in high-stakes certification exams. In our case, we have specialized in exams such as the Bilingual Target Language Proficiency Test (BTLPT) in Spanish and the Bilingual or ESL Supplemental.

One of the ways we are innovating in this space is through AI-powered tutoring, which has become a central area of specialization for us. That’s why we created Verónica, our AI tutor, but Verónica is more than a chatbot. Yes, students can ask her questions in a chat window, but the real power lies in how she is built and used in our online preparation course and hybrid professional development workshops. Instead of relying only on a general language model, Verónica is connected to retrieval-augmented generation (RAG) systems that draw from curated databases on best practices for ESL classrooms, strategies for working with multilingual learners, bilingual education, and the competencies required for teacher preparation and certification exams. On top of that, we use embeddings to strengthen her command of academic vocabulary and to refine the quality of her responses. This means that when students interact with Verónica, her feedback is not only fluent but also grounded, accurate, and aligned with the real expectations educators and future teachers face.

From Scenarios to Smart Feedback

Rather than offering open-ended chats, we’ve focused on building scenario-based learning activities. These are short, realistic tasks (just like those found on teacher certification exams) that learners complete inside our courses and professional development workshops.

For example, a pre-service teacher might listen to an audio prompt in Spanish, such as a parent-teacher conversation, a science lesson, or a community issue, and then record their response. That response is automatically evaluated using AI, which provides structured feedback aligned with the same criteria human evaluators use. This approach transforms practice from something abstract into an experience that feels authentic and targeted.

We want to emphasize that AI is not meant to replace authentic classroom preparation. Teaching is, and always will be, a human practice. What our system does is help teachers make sense of theory in practical ways through scenario-based learning. And when a learner is struggling with a particular concept, competency, or scenario, the system can flag it so that a human expert steps in. For pre-service teachers working in universities or in our online courses, this might mean sending an alert to an instructor at their institution or connecting them directly with one of our subject matter experts for personalized feedback or even a Zoom conversation. For in-service teachers in a school district, flagged challenges can be shared with instructional coaches, mentoring teams, or other district support staff who can provide targeted guidance. In both cases, the AI serves as a bridge, not a substitute, ensuring that timely practice is paired with expert human support at every step of teacher preparation and professional growth.

Getting Technical Under the Hood

For those curious about the technology, here’s a glimpse under the hood. In our courses, Verónica often appears first as a chatbot that can answer questions about the course itself, its objectives, goals, and expectations. She isn’t just pulling from a general model; she uses retrieval-augmented generation (RAG) to access a curated database that contains all the relevant course information. This means her responses are accurate, contextual, and always tied back to the world of bilingual and ESL education. But even when learners are in a chat, if they ask something broad or unrelated, Verónica has been trained with guardrails that bring the conversation back into the context of teaching multilingual learners and the classroom.

But the most powerful use of our AI tutor comes in scenario-based learning. Here, we generate realistic scenarios based on specific competencies that pre-service and in-service teachers need to demonstrate when working with bilingual or ESL students. The AI is guided by a structured request, which we send in JSON format. For those less familiar, JSON (JavaScript Object Notation) is a way of organizing information into clear, machine-readable fields, giving the AI a checklist of what to look for and how to respond. By formatting both the request and the expected answer in JSON, we can define exactly what is needed: the competency being assessed, the evaluation criteria, the scoring scale, and the structure of the feedback. This ensures that instead of receiving long, unstructured text, we get responses that are consistent, pre-formatted, and easy to store, track, and compare across learners. Because these instructions are anchored in our database (and supported with embeddings to refine vocabulary and linguistic accuracy), the AI’s feedback is not only accurate but also aligned to professional standards and directly useful for teacher growth.

Every response (oral or written) is stored (with user consent), which allows us to build personalized learner profiles. Over time, the system “remembers” where each teacher left off, tracks their progress across competencies, and adapts the next scenarios accordingly. This ensures that practice is not only dynamic and authentic but also part of a longer growth journey.

Why This Matters

This combination of bilingual education expertise, technical coding, and database integration is what makes our approach unique. We aren’t satisfied with surface-level AI tools. We want to build systems that:

• Respect the rigor of professional teacher preparation.
• Provide feedback that actually helps students grow.
• Adapt to where each learner is in their journey.
• Keep human experts in the loop, ensuring that AI is guided, checked, and enriched by the knowledge of experienced educators and specialists.

Our work with our AI tutor and our scenario-based activities shows that AI, when carefully designed, can do far more than answer questions. It can guide, evaluate, and support growth in ways that scale across schools, universities, and entire professional communities. Most importantly, the system itself acts as a bridge between learners and human experts, flagging moments when personal guidance is needed and making sure that educators remain central to the process.

And this is just the beginning.

*********************************

Below you will find a short explanatory video that we created using the same tool we’ve been using since last year to generate podcast versions of our blog posts: NotebookLM from Google. Recently, Google released major upgrades to NotebookLM, including the ability to create video overviews of provided content (see announcement here). We asked the tool to generate a short video explaining why we believe our use of AI tutors (like Verónica) is a powerful approach for teacher preparation. What’s remarkable is that it only took one attempt, and the AI produced a clear and engaging explanation. We invite you not only to watch the video but also to consider how this new feature of NotebookLM could be used in your own classrooms.