(Written by Ry, a TTS-built custom Copilot AI Chatbot. Edited by Andrew West, Freedom Baird, and Justin Horvath)
 

What would it look like if a course had its own AI tutor—one that knew the lectures, the syllabus, the instructor’s voice, and, crucially, when not to answer certain questions? For Andrew West, professor in the Department of Chemistry and the Tufts Prison Initiative, that question became a semester‑long experiment in his large interdisciplinary course Big Bang to Humankind during spring 2026. The result was ChatBBtHk, a custom generative AI chatbot designed to help students review course material and generate adaptive practice problems that increase in difficulty using Bloom’s taxonomy. 
 
West is known for taking an innovative approach to teaching, often partnering with students to test new instructional methods. ChatBBtHk grew out of that same mindset. Rather than positioning generative AI as a shortcut, West treated it as a designed teaching tool—one with clear boundaries and a specific pedagogical purpose.  
 
West built ChatBBtHk during his participation in the pilot of the Playlab.ai tool, which allows users to design and share custom AI apps. During the pilot, offered and supported by Tufts Educational Technology Services team, West trained his chatbot on lecture transcripts from multiple semesters, course slides, Piazza posts, the course syllabus, and past assignments, creating an AI that was well contextualized to the course rather than being a general‑purpose model.  
 
At the heart of the project was a simple but powerful idea: practice matters, and not all practice is equal. ChatBBtHk was designed to generate review questions and problems that could scale in cognitive complexity as students interacted with it. Early questions might focus on recall or basic comprehension, while later prompts pushed students toward application, analysis, and synthesis—mirroring the progression outlined in Bloom’s taxonomy. As students demonstrated understanding, the chatbot increased the level of challenge, encouraging deeper engagement with course concepts. In practice, this meant that a student studying stellar evolution, for example, might move from identifying basic properties of stars to reasoning through how stars of different masses would evolve over billions of years—questions that require conceptual understanding rather than flashcard‑style recall. 

Importantly, West also spent significant time teaching the limits of the tool. ChatBBtHk was explicitly trained not to answer homework or quiz questions, and students were told from the outset when AI use was appropriate and when it was not. West demonstrated the tool in class on the day it launched, linked to it directly in Canvas, and invited ongoing student feedback—both informally and through a required survey later in the semester. That openness extended to ethics as well: students were allowed to opt out of AI use entirely on ethical or pedagogical grounds and still receive full credit, provided they explained their reasoning. Fourteen students chose that option, citing concerns ranging from environmental impact to cognitive dependence. 
 
Student use of the tool was substantial. In a class of 94 students, 80 used ChatBBtHk at least once during the semester, generating hundreds of sessions and thousands of messages. Usage spiked predictably before exams, with some students using the chatbot to generate large sets of practice problems that functioned like personalized practice exams. Survey responses at the end of the term reflected generally positive experiences, with students praising the bot’s accuracy, its grounding in course materials, and its refusal to cross boundaries into answering graded questions directly.  
 
At the same time, West has been candid about the tradeoffs. ChatBBtHk could be verbose. It sometimes struggled to follow very specific instructional constraints. And its presence coincided with a sharp drop in office‑hour attendance—something West sees as a potential concern rather than a win. Perhaps most tellingly, when West required students to use AI on one exam question—deliberately choosing a question the AI would get wrong—many students trusted the chatbot’s answer rather than questioning it. The moment became an important lesson in AI literacy: tools can be helpful, but they still require skepticism, domain knowledge, and human judgment. 
 
For faculty and instructors curious about generative AI but unsure where to begin, ChatBBtHk offers a useful case study. It shows how AI can be shaped around pedagogical goals, how adaptive practice can be designed to support deeper learning, and how transparency and boundaries are essential when using a new teaching tool. 
 
To learn more, visit Tufts AI and Teaching guides or email Tufts Educational Technology Services for a one-on-one consultation.