By Scott Virkler
Hawkes Learning’s AI Tutor is a pedagogically designed, student-centered learning companion that meets students where they are on their learning journey. At the point of need, the AI Tutor provides learning support and tips in an encouraging way for the student in both English and Spanish.
We are a small company and compete with large companies with seemingly endless resources. We had a very small team working on the creation of this product with a limited budget. The limits placed on the team forced us to focus on solving the most important problems first and not be distracted with niceties that did not enhance learning. We used a commercially available Large Language Model but limit the data set being used to Hawkes content for each specific course. The solution was also built so we can easily vary or switch the model we are using.
We engaged instructors and students early in the development process. Because there is still a good bit of negativity about the use of AI in Education for students, we knew it was important for the instructors to not only help design the approach we were using with AI but also encourage the students to use it. The instructors’ input was invaluable. Without their feedback, it would have been impossible to produce the quality of AI tutor we were seeking to develop. This early engagement helped us avoid unnecessary mistakes that would have distracted us at a critical time during the launch.
As generative AI was picking up steam in the market, we focused on what makes learning happen. AI Tutor was built around learning research-based best practices:
Contingent Scaffolding: The best tutors respond to student input in real time, adjusting their help based on what the learner says or does. AI Tutor mimics this by adapting to each student’s needs mid-conversation.
Zone of Proximal Development: Students learn best when they’re challenged just beyond what they can do independently. AI Tutor helps them work through a problem without giving away the answer—promoting real understanding.
Retrieval Practice: Asking students to actively recall information helps them retain it better than re-reading or reviewing. AI Tutor builds this into its prompts to strengthen long-term memory.
Self-Determination Theory: Motivation matters. Encouragement, autonomy, and choice boost student persistence. AI Tutor reinforces effort, celebrates progress, and creates a low-pressure, high-reward learning space.
Cognitive Load Theory: Learning is harder when the interface is cluttered or overwhelming. We intentionally designed the UI to be clean, focused, and paced to support working memory.
Metacognition: Students become better learners when they reflect on how they’re thinking. AI Tutor prompts students to explain their reasoning and revise their approach, creating a thoughtful and intentional learning experience.
What we learned
We continue to learn significant lessons with this product. We learn from students, instructors, our internal team, and the tool itself.
The limited data set not only minimizes hallucinations but also helps to ensure greater accuracy in the responses. We learned that Prompt Engineering (the process of defining and refining inputs to the AI) is critical. When done well, the AI can work great, but if off at all, or if the underlying model changes, the Prompt Engineering must be revisited to maintain accuracy.
AI has a mixed reputation among educators, and we were prepared for pushback. With that in mind, we designed AI Tutor to be different. Its intentional design focuses on developing problem-solving skills, building confidence, and becoming independent learners—benefits that extend far beyond the classroom.
One challenge with designing AI Tutor was figuring out how to ensure students are getting relevant answers also aligned with our courseware and teaching pedagogy. We were able to overcome this by implementing thoughtfully crafted guardrails to ensure guidance is always consistent, aligned, and accurate.
AI was created to provide answers, so we essentially had to break the AI. We learned how this is done varies by the underlying subject matter. We had to have the AI give bite-size responses. This aspect of the creation of AI Tutor was challenging because AI tends to be long-winded or tries to answer far beyond what is needed for a guided learning experience. Although difficult, this was a crucial decision in creating the tool, so students are engaged and continually ask questions in their learning.
We knew student needs must be at the forefront of the design. We didn’t just assume what we thought instructors might like. Feedback showed:
- They feel comfortable asking questions they might avoid in class.
- They want to learn, not use AI to take shortcuts.
- The tool works where students spend the most time (Practice mode) and is easy to use.
- Instructors support using AI Tutor once they see it provides consistent, aligned support while preserving academic integrity; supporting our notion instructors aren’t against AI… they’re against it being used in the wrong way.
We built the Hawkes Learning platform to help students learn, differing from the standard approach which is to have homework platforms built to supplement a textbook. This subtle difference is critical regarding how students view and use our learning platform. We strive to approach education through a research-backed, mastery-based lens—focusing on building understanding, confidence, and long-term learning rather than merely completing tasks. One way we’ve done this historically is through our “Explain Error” feature that provides immediate, error-specific feedback, helping students understand not just what was incorrect, but why and how to correct it. This method reinforces critical thinking, deep comprehension, and problem-solving skills.
AI Tutor takes this same principle and builds upon it—when students interact with AI Tutor, it doesn’t hand out answers or shortcuts. Instead, it understands the context of the learning concept the student is studying, the specific question they are working on, the student responses, and identifies misunderstandings in real time to meet them where they are.
In addition, it offers hints, explanations, and targeted questions that mirror the scaffolding an instructor would provide during remediation or office hours and encourages students to reflect on their reasoning, revise their approach, and arrive at the correct solution themselves. Every learner has access to the same pedagogically designed guidance they could expect from a human tutor, but at scale, and accessible anytime, anywhere.
The limited data set approved by the instructor ensures students are studying and practicing on the specific learning concepts the instructor wants them to focus on. Other AI learning tools are using much broader data sets which not only increase the risk of hallucinations but also create distractions for the students.
How it is working out
While there continues to be much press about AI and learning, we found what seems to be a logical use of AI for both instructors and students. The response from instructors and students has been overwhelmingly positive. Instructors love that the AI Tutor doesn’t give away the answers, and is specifically focused on the learning concepts they have selected to teach. Students like the encouraging feedback, pre-selected concepts to pick from, tone of the tool, and that it is in just the place they need it – when they need it. Both tell us they appreciate the tool is available anytime students needs help, day or night.
About the author
Scott Virkler is the CEO of Hawkes Learning, where he champions a culture that values collaboration, positivity, and continuous improvement. He has a passion for education and empowers teams to solve meaningful challenges and make a greater impact on learners everywhere. Under Scott’s leadership, the company continues expanding its suite of cutting-edge tools—all with the goal of making education more accessible, equitable, and effective.
