SimplrAI: Engineering Grade-Appropriate AI Explanations to Reduce Cognitive Load and Improve
Advisor: Jonathan Bielli
Internship Location: Terno AI
Onsite Mentor: Sandeep Giri
Artificial intelligence is rapidly becoming a default academic tool for students, yet most large language models were not designed with pedagogical clarity, grade-level appropriateness, or cognitive simplicity in mind. As a result, students often receive explanations that are verbose, poorly structured, or mismatched to their reading level—paradoxically increasing confusion rather than improving understanding. My project addresses this gap by designing and evaluating SimplrAI, an AI-based academic explanation system that integrates structured prompt templates, grade-level controls, and readability scoring tools. This project tests the hypothesis that explanations engineered with intentional structure—definitions, stepwise reasoning, examples, and applications—will significantly improve perceived clarity and learning outcomes compared to default AI responses. By combining computer science, cognitive load theory, and user-experience design, SimplrAI contributes a replicable methodology for evaluating AI explanations in real educational settings.
Project Blog
Post 1: The Problem I Set Out to Solve — And Why It Matters
I didn't start SimplrAI because I wanted to build another AI app. I started it out of frustration.
Contrary to what I heard, senior year got heavy fast. Multivariable calc, data structures, dense research papers, lecture slides stacked on top of each other. I was staying on top of notes and readings, doing everything you're supposed to do, but I kept feeling like I wasn't actually understanding things at the level I wanted to. It took me a while to figure out why, and when I did it was kind of obvious in hindsight.
My notes were in one place, practice problems in another, and explanations spread across random YouTube videos and textbook pages I barely remembered saving. I was spending more energy trying to keep track of everything than actually learning from it.
So I started building SimplrAI with the guidance of Sandeep Giri. The idea was pretty simple at its core: instead of just storing information, what if something could help you actually process it? Break it down, surface what matters, and push you to think through ideas rather than just transcribe them. When I brought it up my fellow students, the response was pretty unanimous. Everyone felt the same: there was just too much content coming at them with no real system for making sense of it.
That's what SimplrAI is trying to solve. Not access to more information, because students already have plenty of that. Just a smarter way to actually learn from it. In the coming weeks, I'll be testing and refining SimplrAI to create an educational AI that really works.
If you want to check it out, the link is simplrAI.tech And if you have any feedback after trying it, I'd genuinely love to hear it: https://forms.gle/vmSW3U2DxReuEia4A
Week 1: Building the First Features of SimplrAI
This week was the first real development phase for SimplrAI. Instead of just thinking about the idea, I focused on building the initial features that would form the foundation of the platform. My goal for Week 1 was simple: create a basic system where students can input notes and receive clearer, structured explanations in return.
The first feature I implemented was a note input system. This allows a user to paste raw notes, lecture text, or copied material into the platform. The idea is that students often have messy or unorganized notes, so SimplrAI needs a simple way to accept that raw information as the starting point.
After that, I built the AI explanation generator. This feature takes the input notes and processes them to produce a simplified explanation of the topic. Instead of repeating the notes word-for-word, the system restructures the information into clearer language so that the core concept is easier to understand.
I also added a key concept extraction feature. When notes are submitted, SimplrAI now identifies the most important ideas and outputs them as a short list of core concepts. This helps students quickly see what actually matters in a large block of information.
Another small feature I added was structured output formatting. Instead of returning a large paragraph, the system now organizes responses into sections like simplified explanation, key concepts, and quick summary. This makes the output easier to read and more useful for studying.
By the end of Week 1, SimplrAI can take raw notes, analyze them, and return simplified explanations along with key concepts. It’s still an early prototype, but these features form the basic workflow that the rest of the platform will build on.
Week 2: Expanding Inputs and Improving the Workflow
During Week 2, I focused on expanding what SimplrAI can actually accept as input. In Week 1, the system mainly worked with pasted notes. That was a good starting point, but it wasn’t realistic for how students actually gather information. Most of the time, content comes from multiple sources—web pages, documents, lecture slides, and other materials. So this week was about making SimplrAI more flexible.
One of the biggest features I added was link-based content extraction. Now, instead of manually copying text, a user can paste a web link and SimplrAI will extract the main content from that page. This allows students to pull information directly from online articles, course resources, or research pages and process it through the system.
I also implemented Word document support. Students often download lecture notes or assignments as .docx files, so SimplrAI can now read those files and extract the text automatically. Once the content is extracted, it goes through the same processing pipeline as regular notes—generating simplified explanations and identifying key concepts.
Another improvement I worked on was cleaning and structuring the extracted text before it goes into the AI processing stage. Raw web pages and documents often contain extra formatting, headings, or irrelevant content, so I added a step that filters and organizes the text before analysis. This makes the explanations generated by SimplrAI more accurate and easier to follow.
By the end of Week 2, SimplrAI became much more practical to use. Instead of only working with manually typed notes, it can now pull information directly from links and documents, which moves the platform closer to how students actually study and gather information.
Week 3: Making SimplrAI More Interactive
Week 3 was less about adding completely new core features and more about making SimplrAI feel like a real and usable system. I focused on improving structure, adding interactivity, and planning the next layer of learning tools
One of the main updates this week was improving the overall output structure and clarity. In earlier versions, the community feedback said that responses were helpful but not always consistent for learning. I refined how SimplrAI organizes explanations so the every outputs now follows a clearer format –– simplified explanation -> key concepts + concise notes -> overall summary. This chronological format makes it much easier to read and actually use while studying.
I also started working toward PDF support, which has been one of the most requested features so far. A lot of academic material –– textbooks, worksheets, research papers –– come in PDF format, so integrating this is important for making SimplrAI more practical. While full support isn’t completely implemented, the extraction pipeline is mostly complete and will be deployed soon.
The biggest addition this week was introducing a chatbot feature. Instead of just generating static explanations, users can now ask follow-up questions about their material. This makes the experience more interactive and closer to how students fundamentally learn –– by asking questions to clarify doubts and go deeper into topics.
Along with that, I’ve started thinking about a guided learning approach. Instead of just presenting information, SimpleAI should also help users actively engage with it as well.
That is why looking ahead, the next step will be to implement quizzes and active recall tools. The goal is to move beyond short term understanding into long term retention –– making sure that users can actually apply what they learn for a long time.
By the end of Week 3, SimplrAi is no longer a tool that just gives an explanation –– its starting to become a real tool that you can interact with to further your learning and curiosity.
Week 4: Expanding Capabilities and Improving Reliability
Week 4 was a big step forward in SimplrAI’s capability and reliability.
The biggest update this week was fully implementing PDF support. Users can now upload PDF, like textbooks, class notes, or worksheets, and SimplrAI will extract and process the content just like any other input. This was important because a huge portion of academic materials exists in PDF form, and without this, the platform would not be ready for real use.
I also made major improvements to the chatbot system. Earlier, it could answer questions, but the responses weren’t always grounded in the actual material provided. This week, I worked on reducing hallucinations by making the chatbot rely more strictly on the user’s uploaded content. Now, its answers are far more accurate and tied directly to the notes or documents, which makes it much more trustworthy as a reliable study aid.
Another new direction I started working on is video support. A lot of students learn from lectures and Youtube videos, so the goal is to allow SimplrAI to extract and process information from video content as well. This is still in progress, but the early groundwork has been laid.
On top of that, I’m preparing to begin data collection soon. As more users interact with SimplrAI, collecting feedback and usage data will be critical for improving the system––especially in refining explanations, question generation and overall learning flow.
At the end of this week, SimplrAI’s capabilities have expanded far beyond text and documents into a broader learning platform. More importantly, it’s becoming more reliable, which is essential if students are going to trust it as part of their preparation workflow.
Weeks 5-7: Real Users and What Actually Works
Weeks 5 through 7 were less about adding big new features and more about seeing how SimplrAI holds up with real students. Up until now, most decisions were based on what I thought would be useful. These weeks were about testing that.
I started a basic data collection process to understand how people actually use the platform. I looked at what types of inputs they prefer (notes, PDFs, links), how often they use the chatbot, and whether they engage more with explanations or quizzes. I also paid attention to where things break: confusing outputs, unclear explanations, or points where users just stop using it.
To get a better range of feedback, I tested SimplrAI with students in both the United States and India. The usage patterns were a bit different. Students in India used it more for exam prep and direct concept breakdowns, while students in the U.S. used it more for understanding lectures and assignments. But in both cases, the most used features were the simplified explanations and structured outputs.
Alongside this, I made a few targeted improvements. I refined the quiz generation so questions actually match the material better. I also improved the chatbot’s ability to stay within context, especially during follow-up questions. Another small but useful change was improving the multi-source workflow, so users can combine PDFs, links, and notes more smoothly instead of treating them separately.
Most of the work during these weeks wasn’t that cool or fun, but it made a difference. Small fixes to formatting, clarity, and response accuracy made the platform easier to use overall.
By the end of Week 7, the project started to feel different. It’s now shaped by real usage and feedback from students, rather than just my initial ideas.
Weeks 8-9: – Analyzing the Data and Finding Patterns
Weeks 8 and 9 focused mainly on analyzing the data collected from students using SimplrAI. After spending the previous weeks testing features and gathering feedback, I wanted to better understand how students actually study, what they struggle with most, and which parts of the platform were genuinely useful.
One of the biggest patterns I noticed was the difference between student usage in the United States and India. Students in India generally uploaded much larger amounts of material at once –– full chapters, long PDFs, and dense exam review notes. Many of them used SimplrAI during high-intensity study sessions, especially for competitive exams or cumulative tests. Their focus was usually speed, retention, and breaking down large amounts of information quickly.
Students in the United States, on the other hand, tended to use the platform more consistently over time. Instead of uploading huge batches of content, they often worked with lecture notes, homework material, or smaller class-based assignments. Their usage was more centered around understanding concepts gradually rather than rapid review.
Another interesting finding was how users interacted with explanations. I originally assumed longer, detailed outputs would be more valuable, but that was not always true. Most students preferred shorter, structured explanations with clear summaries and highlighted key concepts. Simplicity and readability mattered more than length and depth.
I also noticed that the chatbot became one of the most-used features because students naturally wanted to ask follow-up questions instead of just reading static explanations. This reinforced the idea that learning works better when it feels interactive.
These weeks helped me understand that the next step in developing SimplrAI is about adapting to different learning habits, academic systems, and study behaviors.