Smart Study Buddy

• 

  Console home

• 

• 

• 

  Test

• 

• 

  Amazon Bedrock Agent

• 

• 

• 

• 

• 

  Current bitcoin price

• 

  Requeriment

Inspiration

As a developer passionate about education, I witnessed how 41 million Spanish-speaking students in the US struggle with English-only educational tools. I saw students with homework photos getting no help from traditional chatbots. I realized that existing AI tools require too many manual commands students must explicitly say "search this" or "translate that" creating friction that stops learning.

I wanted to build something different: an autonomous AI tutor that thinks for itself and removes barriers to education.

What it does

Smart Study Buddy is an autonomous AI tutor powered by AWS Bedrock Agents that intelligently orchestrates multiple specialized services:

• 🖼️ Image Analysis: Upload a photo of a math problem, and it automatically detects, analyzes, and explains it step-by-step using Claude Sonnet 4.5 v1's vision capabilities
• 🔍 Real-time Web Search: Ask about recent developments, and it searches the web automatically without being told
• 📰 News Fetching: Inquire about current events, and it retrieves the latest information
• 🌐 Bilingual Support: Write in Spanish, and it responds fluently in Spanish with proper educational terminology
• 🧠 Direct Knowledge: For stable topics, it uses Claude's extensive knowledge base

The agent autonomously decides which tools to use no buttons, no menus, no manual commands.

How I built it

Architecture:

• Frontend: React + Vite deployed on Vercel with responsive design
• API Layer: AWS API Gateway with CORS and request validation
• Orchestration: AWS Bedrock Agent (ID: WYXZJ5YMKT) using Claude Sonnet 4.5 v1
• Backend: 6 AWS Lambda functions in Python (3.11/3.12):
  • Main API handler with retry logic
  • Image analysis with Claude Vision API
  • Web search integration (SerpAPI)
  • News fetcher (NewsAPI)
  • Translation (Amazon Translate)
  • Agent invoker

Key Technical Decisions:

1. sessionAttributes for image storage: Solved data truncation issues by storing base64 images in sessionAttributes instead of parameters
2. Cross-region inference profiles: Used us.anthropic.claude-3-5-sonnet-20241022-v2:0 for better availability
3. Exponential backoff retry logic: Built robust error handling after experiencing the AWS US-EAST-1 outage during development

Autonomous Agent Instructions: I carefully crafted explicit instructions telling the agent exactly when and how to use each tool, enabling true autonomous behavior.

Challenges I ran into

1. AWS US-EAST-1 Outage (October 20, 2025)

During active development, AWS experienced a major outage. This forced me to implement comprehensive retry logic with exponential backoff, making the system production-ready.

2. Image Data Truncation

Bedrock Agent truncated base64 image data to ~16 characters when passed as parameters. Solution: Store images in sessionAttributes where they aren't truncated.

3. Agent Not Calling Image Tool

The agent didn't automatically detect uploaded images. Solution: Added explicit step-by-step protocols in agent instructions with "MUST" and "NEVER" keywords.

4. Rate Limiting

Hit the 2 requests/minute quota repeatedly during testing. Solution: Implemented retry logic and user-friendly error messages while requesting quota increase from AWS.

5. Base64 Data Corruption

Whitespace and quotes corrupted image data. Solution: Comprehensive cleaning pipeline that strips all invalid characters.

Accomplishments that I'm proud of

✅ Production-ready system - Fully deployed and accessible at (https://smart-study1.vercel.app/).

✅ True autonomous orchestration - Agent makes intelligent decisions without user commands

✅ Robust error handling - Survives outages, throttling, and edge cases gracefully

✅ Multi-modal intelligence - Seamlessly handles text and images

✅ Bilingual education - Breaking down language barriers

✅ Cost-effective - Operating at ~$5/month with AWS credits

✅ Real impact - Solving genuine educational inequality problems

What I learned

Technical:

• AWS Bedrock Agents require explicit, step-by-step instructions for autonomous behavior
• sessionAttributes are superior to parameters for large data
• Cloud services need retry logic and graceful degradation
• Comprehensive logging (especially data sizes) accelerates debugging
• Infrastructure abstractions have limits understanding underlying systems is crucial

Product:

• Autonomy dramatically improves UX
• Multi-modal capabilities expand use cases significantly
• Language support isn't optional it's essential for equity
• Real-time information keeps educational tools relevant

Process:

• Start simple, iterate to add complexity
• Test early and often
• Documentation during development pays off
• Adversity (like AWS outages) can improve the final product

What's next for Smart Study Buddy

Short-term:

• Voice input/output for accessibility
• Document processing (.pdf, .docx) for homework help
• Learning analytics dashboard
• User authentication and personalization

Long-term:

• Multi-language support beyond English/Spanish
• AR/VR immersive learning environments
• AI tutor marketplace for specialized subjects
• Integration with school learning management systems (Canvas, Blackboard, Moodle)

Vision: Democratize access to intelligent, personalized education for every student, everywhere.

---

Live Demo: smart-study1.vercel.app
GitHub: github.com/Delis99/smart-study
Video: youtu.be/qs4TNZsxK44

Built With

• amazon-translate
• aws-api-gateway
• aws-bedrock-agents
• aws-iam
• aws-lambda
• aws-secrets-manager
• claude-3.5-sonnet-v2
• claude-sonnet-4.5
• cloudwatch
• newsapi
• python-3.11
• python-3.12
• react-18
• serpapi
• tailwind
• vercel
• vite