Inspiration
Traditional learning platforms often promote passive content consumption, where students read or watch without deeply understanding concepts. This led me to explore the Socratic learning method, which focuses on guiding learners through questions rather than directly providing answers.
With the rapid advancement of Generative AI, I saw an opportunity to build a system that acts as a personalized AI tutor, encouraging critical thinking and active learning instead of memorization.
What it does
NovaMentor is a GenAI-powered learning assistant and agent system that:
Provides interactive AI tutoring through guided questioning
Breaks down complex concepts into simpler parts
Generates adaptive quizzes based on user understanding
Tracks learning performance through dashboards
Encourages active learning with hints, retries, and step-by-step reasoning
Instead of giving direct answers, it helps users arrive at answers on their own.
How I built it
Frontend: Developed an interactive interface for chat, quizzes, and dashboards
Backend: Built using Python with REST APIs to manage AI interactions
AI Integration: Used AWS Bedrock with Nova models
Primary model: amazon.nova-micro-v1:0 (fast and cost-efficient)
Fallback model: higher-capability model for complex queries
Key implementations include:
Model fallback strategy to ensure reliability
Prompt engineering tailored for Socratic questioning
Learning analytics to track performance trends
Logging and debugging mechanisms for stability
Challenges I ran into
Repetitive or generic AI responses Solved by refining prompts and structuring outputs
Latency issues with larger models Addressed using a fallback model strategy
Designing effective Socratic prompts Required multiple iterations to ensure meaningful questioning
Backend errors and integration issues Resolved through systematic debugging and logging
AWS Bedrock configuration and cost management Managed using billing alerts and optimized model usage
Accomplishments that I'm proud of
Built a functional GenAI-powered tutoring system
Successfully applied the Socratic learning approach in an AI application
Designed and implemented a multi-model fallback architecture
Integrated learning analytics with AI-generated insights
Handled real-world debugging and deployment challenges independently
What I learned
Practical implementation of LLMs using AWS Bedrock
Importance of prompt engineering in controlling AI behavior
Designing AI systems that guide thinking rather than just answering
Debugging and handling real-world system failures
Balancing performance, cost, and accuracy in AI applications
What's next for NovaMentor
Add voice-based interaction for a more natural learning experience
Develop subject-specific intelligent agents
Introduce multi-user classroom analytics
Integrate AI with cybersecurity learning modules
Improve scalability and optimize system performance
Built With
- amazon-web-services
- and-fastapi-for-creating-backend-rest-apis.-the-ai-layer-is-powered-by-aws-bedrock-using-nova-models-(amazon.nova-micro-v1:0)-with-prompt-engineering-techniques.-cloud-services-include-aws-(bedrock-and-iam-for-secure-access)
- bedrock
- boto3
- css
- fastapi
- github
- html
- iam
- javascript
- json
- logging
- nova
- promptengineering
- python
- react
- rest
- sqlite
- streamlit
- streamlit-for-building-the-interactive-ui
- with-boto3-used-for-model-integration.-data-handling-is-managed-using-lightweight-storage-(json/sqlite)
- with-html-and-css-for-the-frontend-interface
