Why I Built Elara

As a student, I experienced firsthand how frustrating it is to get stuck on a math problem with no one to help. Traditional tutoring is expensive and hard to schedule, and most online resources optimize for answers—not understanding.

I saw the same pattern repeatedly when helping classmates: students didn't need solutions handed to them—they needed guidance through the reasoning process.

When I explored existing AI tools, I noticed a critical gap. Most systems either:

• Jump straight to the final answer, or
• Provide generic feedback disconnected from the student's actual work

None of them could look at a student's handwritten steps and respond with context-aware guidance.

I set out to build the tool I wished I had:

An AI tutor that watches you think, not just what you type.

Elara is built around the Socratic method—guiding students with the right questions at the right time, instead of giving away answers. The goal isn't correctness—it's building intuition.

To make this possible, the system needs to:

• Understand messy, real student input (handwritten math, incomplete steps, corrections)
• Track reasoning across multiple steps
• Decide when to intervene and how much help to give

Building Elara forced me to go beyond just using LLMs—to designing a full agentic system that reasons over user work.

Some of the core challenges I tackled:

This project changed how I think about building software.

The hardest problems weren't just technical—they were about:

• Defining what "good help" actually looks like
• Deciding when not to intervene
• Designing systems that adapt to messy, real human behavior

I also learned how to:

• Break down ambiguous problems into testable components
• Iterate quickly with real users instead of over-engineering
• Build systems where multiple AI components collaborate toward a goal

Elara is still early, but the goal is clear:

To build a system that becomes a long-term learning partner, not just a homework helper.