Personalized Student Performance Recommendations for NEET Testline 🚀

This project is my solution to provide personalized study recommendations for students using the NEET Testline app. I've built a Python script that analyzes quiz performance data and leverages the power of Google's Gemini AI to deliver actionable insights and recommendations, helping students focus their preparation effectively.

[Video Demo]

Data Overview

I worked with two main datasets to perform this analysis:

• Current Quiz Data: This dataset provides details about a user's most recent quiz submission. It includes information like:

  • Quiz ID and details (topic, difficulty level, etc.)
  • User ID
  • Submission timestamps
  • Score, accuracy, speed, and other performance metrics
  • A response_map linking Question IDs to the selected option IDs.

• Historical Quiz Data: This dataset contains performance data from the last 5 quizzes for each user. It includes:

  • Quiz IDs and details
  • User IDs
  • Submission timestamps
  • Scores, accuracy, speed, and performance metrics for each quiz
  • response_map for each historical quiz.

These datasets are provided via API endpoints.

Task

The core tasks I tackled in this project were:

• Analyze the Data: I explored the schema of both datasets to understand the available information. Then, I focused on identifying patterns in student performance across:

  • Topics
  • Difficulty Levels (if available, currently data has difficulty_level: null)
  • Response Accuracy

• Generate Insights: Based on the analysis, I aimed to highlight key insights for each student, including:

  • Weak Areas: Identify topics where the student consistently underperforms.
  • Improvement Trends: Determine if the student is showing progress over their last few quizzes.
  • Performance Gaps: Pinpoint the difference in performance between strong and weak areas.
  • Speed vs. Accuracy Analysis: Analyze if the student's quiz completion speed is impacting their accuracy.
  • Difficulty Level Recommendation: Suggest appropriate quiz difficulty levels for weak topics.
  • Mistake Pattern (Basic): Count total incorrect answers per topic to find areas of frequent mistakes.

• Create Recommendations: The ultimate goal was to generate personalized and actionable recommendations for students to improve. These recommendations include:

  • Suggested topics to focus on.
  • Recommended quiz difficulty levels for practice.
  • Tips on time management during quizzes (based on speed analysis).
  • General study advice.
  • Specific study actions for weak topics.

• Bonus Points: To add extra value, I also aimed to:

  • Define a Student Persona: Based on performance patterns, assign a creative persona label to characterize the student's learning style.
  • Highlight Strengths and Weaknesses with Creative Labels: Use engaging labels to describe the student's strong and weak areas.

Implemented Features 💪

Here's what I've built into this Python script:

• Data Fetching: I implemented functions to dynamically fetch quiz data from the provided API endpoints using the requests library. Error handling is included to manage potential issues with fetching data.
• Topic-wise Performance Analysis: The script calculates average accuracy, score, and quiz duration for each topic based on historical data, allowing for easy identification of strong and weak areas.
• Performance Trend Analysis: I analyze the accuracy and score trends from the last few quizzes to provide insights into the student's progress over time.
• Difficulty Level Recommendations: Based on average accuracy in weak topics, the script suggests starting with "Easy," "Medium," or "Normal" difficulty quizzes for focused practice.
• Speed vs. Accuracy Analysis: The script analyzes the time spent per question in the latest quiz and compares it to accuracy. It identifies if a student might be rushing and suggests adjusting their pace.
• Basic Mistake Pattern Analysis: I calculate and display the total number of incorrect answers per topic from historical quizzes to highlight areas with frequent mistakes.
• Google Gemini AI Integration: I've integrated Google's Gemini API to generate more human-readable and personalized feedback and study recommendations. The prompt is designed to request structured JSON output for easy parsing and display. The feedback incorporates insights from all the analyses performed.
• Structured JSON Output: The GenAI output is parsed as JSON, providing structured data for feedback, topic-specific recommendations (with actions), and general recommendations. This makes it easy to use the output in an application.
• Text-Based Fallback: The script also includes text-based outputs for all analyses and recommendations, serving as a fallback if the GenAI integration has issues or for comparison.
• Advanced Data Visualizations: Generates insightful visualizations using matplotlib and seaborn, including:
  • Topic-wise Average Accuracy Bar Chart
  • Accuracy Trend Line Chart over Quizzes
  • Quiz Score Distribution Histogram
  • Topic-wise Average Quiz Duration Bar Chart
  • Combined Bar Chart Comparing Weakest and Strongest Topics These visualizations are saved as PNG image files for easy inclusion in reports or dashboards.
• Linting and Documentation: The code is formatted and documented for better readability and maintainability, following Python best practices.

Setup Instructions ⚙️

To run this script, you'll need to:

1. Python Environment: Ensure you have Python 3.9 or later installed.
2. Install Libraries: Install the required Python libraries using pip:

   pip install requests google-generativeai

3. Set Gemini API Key: Obtain a Google Gemini API key from Google AI Studio and set it as an environment variable named GEMINI_API_KEY. For example, in your terminal:

   export GEMINI_API_KEY=YOUR_ACTUAL_GEMINI_API_KEY

4. Run the Script: Execute the Python script from your terminal:

   python main.py

Project Overview and Approach Description 🗺️

This project aims to empower students using the NEET Testline app with data-driven, personalized guidance. My approach is centered around these key steps:

1. Data Acquisition: I start by fetching the necessary quiz data from the provided API endpoints, ensuring robust error handling in case of network issues.
2. Data Processing and Analysis: The core of the script involves processing the historical quiz data to calculate key performance indicators. This includes:
   • Calculating accuracy, score, and quiz duration for each quiz and topic.
   • Identifying strongest and weakest topics based on average accuracy.
   • Analyzing performance trends over the last quizzes.
   • Performing speed vs. accuracy analysis on the latest quiz.
   • Recommending difficulty levels for weak topics.
   • Counting incorrect answers per topic.
3. Insight Generation: The analysis results are then translated into actionable insights, highlighting weak areas, improvement trends, speed concerns, and difficulty level guidance.
4. Personalized Recommendation Generation (GenAI): I leverage the Google Gemini API to create personalized feedback and study recommendations. The prompt I designed provides Gemini with the analysis results (strong/weak topics, trends, speed insight, difficulty recommendations) and instructions to generate:
   • Encouraging and constructive overall feedback.
   • Specific study actions for each weak topic.
   • General study recommendations.
   • The output is structured in JSON for easy use in applications.
   • Generating advanced visualizations (bar charts, line charts, histograms) using matplotlib and seaborn to visually represent topic performance, accuracy trends, score distributions, and quiz durations. The script then saves these visualizations as PNG image files for easy integration into reports.
5. Output Display: The script prints the analysis results, insights, and both the text-based and GenAI-powered recommendations to the console. The JSON output from GenAI is also displayed in a formatted manner.

My approach focuses on combining quantitative data analysis with the natural language generation capabilities of GenAI/NLP to create a truly personalized and helpful learning experience for students.

Bonus Points - Student Persona and Creative Labels 🌟

As a bonus, I have included:

• Student Persona: Based on the observed performance patterns (currently a basic "Inconsistent Achiever" persona), the script assigns a descriptive persona label. This can be further refined with more sophisticated analysis and potentially ML clustering in the future.
• Creative Labels for Strengths and Weaknesses: The script dynamically generates creative and engaging labels for the student's strongest and weakest topic areas, making the feedback more engaging and less dry.

Code Output Screenshot