GeoSolver AI

• 

  Geometry Olympiad Question

• 

  Trying to solve 3d mechanical questions.

• 

  Trying to solves descriptive geometry problems.

• 

  Results of the questions in the video.

• 

  First ai assistance result.

• 

  Problem 1

• 

  Saved Solution of Problem 1

• 

  Background black problems (most of the ai tools cannot understands)

• 

  Saved Solution of Problem 2 (Black Background Problem)

Inspiration

Many students struggle with math and geometry not because they lack intelligence, but because they never fully understand why a solution works. Most tools only provide final answers, leaving learners confused and dependent on memorization.

I wanted to build an AI-powered tutor that focuses on reasoning, explanation, and understanding, not just correctness. GeoSolver AI was inspired by the idea that an AI should think like a teacher, not just a calculator.

How We Built It

GeoSolver AI is built using the Gemini 3 API as the core multimodal reasoning engine.

The system uses an agent-based architecture:

• A Parser Agent interprets and structures the problem
  
• A Solver Agent solves it independently
  
• A Verifier Agent checks each step for mathematical correctness
  
• A Tutor Agent rewrites the solution in a clear, educational style
  
• An AI Assistant Agent generates geometric diagrams and constructions to visually support the reasoning
  

Gemini handles:

• Multimodal input (text + images)
  
• Step-by-step reasoning
  
• Tool-assisted problem solving
  
• Natural language explanations
  

The frontend provides a simple, distraction-free learning experience focused on clarity and learning.

Challenges We Faced

• Ensuring explanations were pedagogical, not just technically correct
  
• Preventing skipped steps in mathematical reasoning
  
• Making the agent system consistent across different difficulty levels
  
• Generating diagrams that actually match the symbolic reasoning
  

We solved these by iterative testing, agent tuning, and prompt refinement.

What We Learned

• Multi-agent systems improve reasoning reliability
  
• Visual explanations significantly increase understanding
  
• Prompt and agent design are as important as model power
  
• AI can act as a thinking partner, not just an answer generator
  

What's Next

• Personalized learning paths based on student performance
  
• Skill tracking and difficulty adaptation
  
• Support for advanced topics like algebra, calculus, and physics
  
• Classroom and teacher dashboards
  
• Mobile-first learning experience
  

Example Math (LaTeX Support)

For example, GeoSolver AI can explain why:

$$ a^2 + b^2 = c^2 $$

represents the relationship in a right triangle using both symbolic steps and geometric visualization.

Built With

• gemini
• html
• typescript