PiPartner 📐🤖

📌 Inspiration

PiPartner was inspired by my high school Precalculus class. The course was tough, and students often had to wait a long time for help. I wanted to build something that could explain math problems clearly, whenever students needed it. My teacher loved the idea and encouraged me to expand it to support IB-style problems too. That’s how PiPartner became an AI-powered tool to provide step-by-step math help and targeted practice.

🤖 What It Does

PiPartner is your personal AI math tutor that supports both text and image input. It can:

• ✍️ Solve typed or handwritten math problems with clear, step-by-step explanations.
  
• 📷 Automatically extract and solve math from uploaded images.
  
• 🧠 Identify the math concept or specific IB topic.
  
• 📚 Generate a set of related practice problems that get harder as you go.
  
• 💬 Keep full conversation memory, so users can revisit or build on past questions.

⚙️ How It Works

🧠 Backend

• Input Routing:

  • If input_type == "text": calls a GPT-4 prompt for concept ID + solution.
    
  • If input_type == "image": sends base64 image to GPT-4 Vision for OCR and solving.
    
  • Each prompt handles both standard and IB-style problems with tailored logic.

• OpenAI Integration:

  • Uses GPT-4 and GPT-4o to:
  • Break down problems step-by-step.
  • Identify IB topics like SL 2.1: Straight-line equations.
  • Generate custom practice questions.

• Conversation Logging:

  • DynamoDB stores:
  • user_id, timestamp, problem, solution, and image (if any).
  • This allows users to return to old problems or ask follow-ups.

• Infrastructure:

  • Flask (Python): Backend logic and routing.
    
  • AWS Lambda: Serverless deployment.
    
  • API Gateway: Frontend-backend connection.
    
  • S3: Placeholder for future image storage.
    
  • UUIDs: Track sessions without user accounts.
    
  • Base64 image checks: Ensures safe and valid inputs.

💻 Frontend

• Framework: Built with React.

• Input Workflow:

  • Auto-detects input type (text or image).
  • Routes to correct backend API.
  • Displays output: explanation + practice problems.

• Features:

  • Clean UI for step-by-step results.
    
  • Re-submission allowed if image input was misread.
    
  • Future-ready design for math keyboard and user accounts.
    

• UX Highlights:

  • Smooth transitions between problem input and explanation.
    
  • Simple error handling for missing or invalid input.
    
  • Designed for clarity — especially helpful for IB math students.

🚧 Challenges

• 🧠 Crafting prompts that worked across a wide range of math topics and levels.
  
• 💸 Managing OpenAI API usage efficiently to reduce cost.
  
• 🖼️ Integrating GPT-4 Vision to both read and solve image-based math.
  
• ⚛️ Pivoting from Flutter to React in a short time.
  
• 🧩 Handling follow-up questions in chat history:
  • Required manual context injection to reattach past problems and ensure accurate responses.
    
  • Built logic to pull previous conversation history from DynamoDB and reformat it for GPT input.

🏆 What I’m Proud Of

• Built and deployed a full-stack AI app from scratch.
  
• Supported both typed and image-based math inputs seamlessly.
  
• Integrated a dynamic chat-like experience with full session memory.
  
• Helped students not only find answers—but understand the "why."

📚 What I Learned

• Prompt engineering is essential for making AI both accurate and helpful.
  
• GPT-4 Vision works well when structured carefully with base64 images.
  
• Managing context in chat-style interactions needs deliberate backend planning.
  
• React is a powerful frontend tool when paired with clean API logic.
  
• DynamoDB is perfect for lightweight, real-time session history.

🔮 What’s Next for PiPartner

• ⌨️ On-screen math keyboard for mobile and accessibility.
  
• 🔁 Follow-up question logic with smoother context handling.
  
• 🧪 More IB/AP-style practice modules.
  
• 📦 User login + saved sessions across devices.
  
• 🌍 Multilingual support (e.g., Spanish, French, Hindi) for global accessibility.
  
• 🧑‍🏫 Educator dashboard for assigning and reviewing student work.

Built With

• amazon-web-services
• claude
• ocr
• openai
• python
• react