Inspiration

I’m an engineering student, and I’ve always believed that we learn by doing, not just by "downloading" information into our brains. True learning happens when we actually train our neural network (our brain).

It's obvious there is a massive gap in how we currently study. When you watch a YouTube video or sit in a lecture, nobody sees how you are thinking. And if your mental models actually match with correct mental model. Even in exams, you get a grade on the final answer, but a correct answer is basically a collection of micro-decisions that we got right.

I wanted to create a tight feedback loop between those thoughts and the results. I wanted a tool that doesn't just grade the final output but actually guides my mental model. I’ve had this idea for a long time, but as a backend guy with zero frontend skills, I never had the time or ability to build this. Until now.

What it does

Here is the core workflow:

  1. The Studio: You can tell the AI what you want to learn (e.g., "Python," "French," "Critical Thinking") or upload a textbook. It builds a custom curriculum path just for you. You can even upload PDFs to make it interactive.
  2. Adaptive Learning: As you solve problems, the system tracks your performance. If you are doing well, the level increases. If you struggle, it eases up or create more variety of question to make your understanding crystal clear.
  3. Two Core Modes:
    • Solo Mode: You solve the question, submit it, and get a mark out of 10 with a helpful tip.
    • Co-pilot Mode: This is the magic. It breaks a question down into steps. It acts like a function loop you give an input, it validates your logic instantly. If you think, "I should subtract 4," and that's wrong, it corrects you right then, ensuring you don't build bad habits.

it’s an engine to learn literally anything. and in any language. It supports over 130 languages.

How we built it

This was my first time truly leaning on AI to bridge my skills gap. I have backend experience, but I don't know frontend.

I used Kiro IDE and their Chat to Code feature extensively.

  • Frontend: The entire UI, the authentication, and the beautiful "vibe" of the site were generated by Kiro. I honestly couldn't have built this without it.
  • Backend: I worked on the logic, but I used Kiro to generate the API endpoints using the spec-driven development feature.
  • AI Logic: The core of Solance uses LLMs acting as a loop. I treat the user as a function: the AI provides an input, waits for the user's output, evaluates the context, and loops again. This allows for that conversational "Co-pilot" feel.

Challenges we ran into

The biggest technical hurdle was taming the non-deterministic nature of LLMs. Since Solance relies on a specific loop where the user is an input function and the AI assesses the output, it was difficult to stop the model from hallucinating or simply giving away the answer.

I spent a lot of time refining the prompts to ensure the AI stayed in "tutor mode" and didn't break the logic flow or skip steps.

On a personal level, my lack of frontend experience was a massive mental block. Creating a platform that actually looks inviting and "fun" to use (rather than just a command line interface) felt impossible at the start.

Accomplishments that we're proud of

Two days ago, I finished the MVP, and I’ve actually started using it to study for my own college syllabus. That is my biggest accomplishment. I’ve built projects before, but I’ve never used one of my own creations as part of my daily life like this.

It genuinely feels more effective than my traditional studying methods. I’m also really proud that I managed to build a full-stack application with a polished UI despite not being a frontend developer. It feels like I unlocked a new superpower.

What we learned

I learned that AI models are definitely not deterministic! A big challenge was prompt engineering making sure the AI acts as a strict tutor and doesn't just hallucinate or give away the answer. I had to refine the prompts to ensure the "Co-pilot" mode actually checks the logic step-by-step.

What's next for Solance

I am going to keep using this for my own studies, so I’m definitely going to keep improving it.

  1. Voice Mode: I want to add a feature where you can just listen to the question and talk to the Co-pilot, making it a hands-free tutor.
  2. Stability: I need to iron out some glitches to handle the non-deterministic nature of the AI better.
  3. Community Paths: I want to make it easier for users to share the courses they created in the Studio with other learners.

I’m really open to recommendations I just want to make learning as fun and effective as possible.

FOR TESTING: email: test@user.com password: abc123

Built With

Share this project:

Updates