Could you share a bit about your career and what inspired you to create this app?
Over the years, I’ve worked as a data engineer at Stord, and a senior data analytics consultant at Kaizen Analytix. Currently, I’m an analytics engineer at Workday. My journey into app development was inspired by my own GRE preparation experience. This test measures multiple skills. I excelled in Quantitative Reasoning but as a non-native English speaker struggled with Verbal Reasoning. It’s hard because you have to master over 1000 words. This personal challenge motivated me to create Scafwording to help others, especially non-native speakers, enhance their vocabulary and achieve better scores.
Interestingly, my academic background also played a pivotal role. During my thesis, I explored the use of reinforcement learning (RL) in a simulated intelligent tutoring system. The goal was to determine the most efficient hint types for human anatomy students. This research laid the groundwork for Scafwording’s adaptive hint system. The techniques I developed then are now applied to GRE vocabulary learning. This proves how life comes full circle: from helping anatomy students with personalized hints to helping GRE aspirants master vocabulary. After all these years I continue to focus on developing technology to help people enhance their learning experience.
How did you identify the need for a new approach to GRE vocabulary learning?
The need for an innovative approach to GRE vocabulary learning became apparent when I was studying for GRE. I realized that simply memorizing words was insufficient for success in the Verbal Reasoning section. Besides, traditional methods often lack personalization to individual learning patterns and effective retention strategies. Students or users are essentially taught the same thing over and over by showing them the same cards and giving them repetitive hints. The traditional approaches also miss contextual understanding of words which is particularly critical when preparing for tests. I wanted to address these gaps and offer everyone getting ready for GRE an adaptive learning experience that goes beyond rote memorization.
What are the key features that set Scafwording apart from other educational platforms?
I would name four innovative features that set Scaffolding apart:
- Adaptive hints. The app employs reinforcement learning to tailor hints based on user performance. That’s how I offer personalized learning experiences.
- Daily quizzes for retention. Quizzes combine previously learned words, tricky words, and missed words to reinforce retention effectively.
- Progress tracking. To encourage consistent learning I implemented streaks and retention scores, so users can monitor their improvement.
- Personalized learning sessions. Each session includes a mix of new words and previously flagged difficult words. Because it’s all about balance!
What technologies and programming languages were used to develop Scafwording?
It’s critical for us to create an efficient and adaptive vocabulary learning experience. That’s why even for the MVP, I utilize a diverse combination of technologies and programming languages. For example, the frontend, data storage and app logic is Bubble—it’s a no-code platform that handles all that. I use Flask, a Python framework, to implement the reinforcement learning model and create API endpoints. I also use PythonAnywhere to host the Python-based backend and machine learning. Bubble’s built-in database is what I use for data storage, user progress, and learning analytics. For API integration, I rely on RESTful APIs to link the Bubble frontend with PythonAnywhere. I also integrated Google Authentication for user login functionality.
What specific reinforcement learning algorithms are employed to personalize hints, and how were they implemented?
Scafwording employs a Q-learning algorithm. This is the most effective way to train an agent on what to do in different situations to get the best results. It’s model-free, so it doesn’t need to know how the world around it works. It figures things out by trying actions, seeing what happens, and learning from the results—even if things don’t always happen the same way. That’s why I use this type of reinforcement learning to personalize the experience.
How did you integrate data tracking, such as answer accuracy, hint effectiveness, and retention, into the app’s architecture?
It was crucial for Scarfwording to integrate Bubble’s database with a custom reinforcement learning model. That ensures a personalized learning experience that evolves with each user’s performance. The system monitors accuracy, adjusts hints based on real-time feedback, and adapts sessions to align with user progress, focusing on challenging words they flag.
I track answer accuracy in Bubble’s database. The users’ responses are collected with their UserID, WordID, Correct (Boolean), and Timestamp. Then Imark correct answers as “learned”; incorrect ones are flagged for review. I also use Bubble’s database to calculate and store retention tracking. Basically it tracks words using the Quiz Performance schema. I calculate the Retention Score—Total Correct / Total Attempted × 100. And then I flag incorrect words for review.
Another important metric for Scafwording is a hint of effectiveness. It’s important to know if we’re going in the right direction. This is managed via API calls to the RL model. For example, when “Don’t Know” is selected, Bubble requests the top three hints from the RL model API. The user chooses a hint, answers a question and gets updated hint rankings, which optimize future selections.
Last but not least—learning progress, which I also manage through Bubble’s database. I track learning status with UserID, WordID, Status (Learned/Reviewed), and Timestamp. Additionally, data such as daily streaks and session duration help us with engagement metrics.
While the app targets GRE takers, how easily can it be adapted for other tests or language learning?
It was very important for us to create Scafwording as highly adaptable. The application can be repurposed for various vocabulary-intensive tests and basically for any language learning purpose. There’s a word database you could easily replace with vocabulary for all the main standardized tests like GRE, TOEFL, IELTS, SAT, GMAT, ACT, as well as specialized vocabulary for professions—medical, legal, business English, you name it. It can also be purposed for different academic disciplines, and, of course, general English language learning.
One of the reasons Scafwording is applicable to various language learning scenarios and different proficiency levels is our hint system. It is based on context, dialogue, and story, which makes the application suitable for different proficiency levels and learning objectives. Also, the Q-learning algorithm is pretty flexible and can be fine-tuned for various test formats or learning goals.
All that being said, with minimal modifications, Scafwording can serve as a versatile tool for vocabulary acquisition across a wide range of educational and professional contexts.
What features or enhancements are you planning for future versions?
I believe Scafwording has huge potential for growth. In the future I see that the product can be enhanced with four key improvements: a user-centric model, expanded hint system, mobile app development, and test expansion.
Currently, the app uses a universal model updated based on all users’ data. Transitioning to a user-centric model would be a significant enhancement for Scafwording. If we implement personalized Q-tables for each user, reinforcement learning will adapt to individual learning patterns. To adjust to individual needs quickly, we’ll have to use techniques like meta-learning or transfer learning. Also, a more detailed profile of a user should be included, taking into account matters such as learning style and prior knowledge. I am sure all of that will lead to a more tailored learning experience. Which means it could increase engagement and knowledge retention.
Our hint system gets a lot of credit, as I mentioned already. The current one includes context, dialogue, and story hints, which are all text-based hints. In the future, we’d love to include image and video-based hints. It’s very important because we all have different approaches to learning and it’s also good for everyone to combine learning styles—visual, auditory, and reading/writing. Additionally, I want to offer more context for complex words and more engaging learning materials.
To make Scafwording more accessible and convenient, I plan to develop a dedicated mobile application for iOS and Android platforms. The key features I want include: seamless syncing across devices, offline access, so users can continue learning even without an internet connection; push notifications with reminders for daily learning to maintain user engagement; and of course a mobile-optimized interface for more intuitive navigation.
And what will the test expansion look like?
Currently, the web application is designed specifically for GRE words. Future versions could be expanded to cover tests where the reading section assesses the user’s ability to understand words in context. Answering questions accurately requires a wide vocabulary. I am talking about very popular tests: TOEFL, SAT, IELTS, ACT, etc.
Reinforcement learning is used to make the process more personalized, as I mentioned earlier. It is based on real-time tracking, personalized learning, and mastery focus. With all these features it adapts learning dynamically. Now that the app concentrates on content, I’d also like to add real-time tracking, and a mastery focus in the future.
The planned enhancements will ensure a dynamic, evolving learning platform for users, which aligns with the latest trends in adaptive learning.
