Inspiration
We were inspired by a common pattern we saw in STEM: students often understand advanced concepts but struggle because foundational misconceptions were never corrected early on. These small errors compound over time, turning into long-standing habits that continue to undermine performance at higher levels.
Because many students feel embarrassed asking questions about concepts they believe they “should already know,” these misunderstandings often go unaddressed the moment they matter most. SecondLook was built to intervene precisely at that moment, stopping incorrect learning before misconceptions reinforce themselves and become barriers to long-term progress.
What it does
SecondLook is a real-time, vision-powered STEM tutor that watches handwritten math work on an iPad and detects conceptual mistakes as they occur. Instead of waiting until the final answer is wrong, it identifies the exact step where reasoning breaks down and intervenes immediately to explain why the approach is incorrect and what the correct next step should be.
At the end of each session, SecondLook generates personalized review notes that summarize recurring misconceptions and highlight what the student should focus on reviewing next.
How we built it
We built SecondLook as a web-based application that connects to a live screen share of a student’s iPad while they work in apps like GoodNotes or Notability. Using real-time vision streaming, handwritten steps are interpreted by a vision-language model and analyzed for incorrect reasoning or invalid rule application.
When a conceptual error is detected, the system triggers an on-screen intervention. At the end of the session, a separate summarization model generates targeted review notes based on the mistakes observed throughout the session.
Challenges we ran into
One of the biggest challenges was connecting all parts of the system into a cohesive real-time pipeline. We had to learn two new platforms, TRAE and Overshoot, while simultaneously developing the MVP and defining a realistic project scope within a short time frame.
Balancing technical ambition with what was actually shippable during the hackathon forced us to make careful tradeoffs and prioritize core functionality. Additionally, merging all our precious work was a very challenging part. Sometimes our merge conflicts would be over 100 lines but through a careful process, we were able to safely merge everything together.
Accomplishments that we're proud of
We’re proud that we built a product with the potential to help a wide range of learners: from younger students just building foundations to advanced students preparing for standardized tests like the ACT, SAT, and math competitions.
Most importantly, we created a system that focuses on how students think, not just whether they get the answer right.
What we learned
We learned how to integrate real-time vision tools like Overshoot with AI development workflows using TRAE, and how to combine multiple models into a single, functional web application.
Beyond the technical skills, we learned how important it is to clearly define scope and design around a specific educational problem rather than trying to build a general-purpose tutoring system.
What's next for NexHacks Project
Next, we plan to roll out SecondLook for free to elementary school students to encourage early adoption and prevent foundational misconceptions from forming in the first place. As students grow familiar with the platform, we plan to expand to middle and high schools, as well as tutoring centers.
In parallel, we aim to partner with SAT and ACT bootcamp providers and college prep consulting companies to offer a dedicated SAT/ACT version of SecondLook with differentiated pricing. This version would allow students to begin standardized test preparation as early as they want, using technology that aligns with the fully digital SAT. For students who prefer not to rely on textbooks or cannot afford traditional one-on-one tutoring, SecondLook provides a more accessible, self-guided alternative. For users whose schools or programs do not partner directly, SecondLook would operate on a subscription-based model.
Because SecondLook does not generate answers or complete problems for students, it aligns more closely with existing academic integrity policies than traditional AI tutoring tools. The system focuses on identifying reasoning errors and reinforcing correct thinking rather than providing solutions. This makes SecondLook easier for schools, test prep organizations, and tutoring centers to adopt as a learning aid rather than a replacement for instruction.
Log in or sign up for Devpost to join the conversation.