Drona AI Tutor
Your AI tutor for smarter, stress-free learning.
Created on 14th April 2025
•
Drona AI Tutor
Your AI tutor for smarter, stress-free learning.
The problem Drona AI Tutor solves
- Shallow Learning Habits
Many students focus on rote learning to pass exams but lack the depth required to truly understand and apply what they’ve learned.
📌 Drona encourages critical thinking by asking follow-up questions, Socratic challenges, and self-reflection prompts.
- Lack of Personalization
Most educational tools are “one size fits all.” Students with different learning speeds, styles, or foundational gaps struggle to keep up.
📌 Drona adapts the conversation and topic flow based on individual understanding and progress levels.
- Overwhelming AI Interfaces
Most AI tools, like ChatGPT, require good prompting skills to get useful answers. For school/college students, this becomes a major barrier.
📌 Drona’s interface is designed to work without perfect prompts—it understands intent, context, and guides the learner naturally.
- Limited Feedback and Practice
Many platforms give content but lack practical assignments, labs, and feedback systems.
📌 Drona includes assignments and lab generation for each level of Bloom’s taxonomy, plus scoring and feedback features.
Challenges I ran into
🧗♂️ Challenges I Ran Into While Building Drona
-
Prompt Engineering for the Socratic Method
Designing effective prompts for Gemini to simulate Socratic questioning was challenging. I had to ensure the AI stayed on-topic, asked the right level of questions, and didn’t just give away answers. -
Maintaining Chat Context Across Sessions
Saving and restoring multi-turn conversations using pickle while linking them to specific users and topics required careful session management and database integration. -
Handling Bloom’s Taxonomy Progression
Implementing level-based learning flow was complex. I had to detect when a student was ready to move to the next Bloom level and prevent skipping or regressions. -
Low-Prompt Design
Creating a system where users didn’t need perfect prompts involved a lot of testing. I had to make Drona intuitive and responsive even with vague or unstructured queries. -
Generating Assignments and Labs Dynamically
Getting Gemini to consistently return well-formatted JSON with valid questions and lab exercises across different topics was tricky and sometimes inconsistent. -
Progress Tracking & Data Storage
Storing and retrieving user progress, assignments, chat history, and levels accurately in SQLite while maintaining performance took thoughtful database design. -
Balancing AI Response Length and Interactivity
AI models can be verbose. I had to tune prompts so Drona responded with bite-sized explanations and maintained engagement without overwhelming the user. -
Security and Authentication
Implementing secure login with password hashing (bcrypt) and managing user sessions safely was critical to avoid vulnerabilities. -
Front-End Integration
Connecting backend features (like topic switching, Bloom-level updates, and assignment APIs) to the frontend UI using JavaScript and Flask templates required clean logic and synchronization. -
Debugging Gemini API Responses
Since AI responses can be unpredictable, debugging Gemini outputs and
Tracks Applied (1)
Software Development
Technologies used