Student AI Workflow

Overview

This project is part of the Microsoft Elevate Internship (December 2025 Batch).
It extends a Power BI student marks dashboard into an AI-powered workflow using Python, Copilot, and Power Automate.
The goal is to automate student performance analysis, generate predictions, and deliver insights through a modular, reproducible pipeline.

---

Features

• Power BI dashboard with KPIs, grade distribution, and subject averages
• Student marks dataset (Excel) for rule-based and predictive analysis
• Python ML pipeline (scikit-learn) for pass/fail and grade prediction
• Copilot-generated insights and conditional formatting logic
• Power Automate workflow for automated reporting and refresh
• Organized folder structure for reproducibility and clarity

---

Tech Stack

• Python (*required) → Download
  Libraries: pandas, scikit-learn, joblib, openpyxl
• Power BI Desktop (*required) → Download

Future Extensions (Optional)

• Microsoft Power Automate → Can be added by users for workflow automation/reporting
• Microsoft Copilot → Can be integrated via Copilot Studio to build AI agents

---

Screenshots

🔹 Run Pipeline [1st]

🔹 Train Model [2nd]

🔹 Marks & Result Analysis Dashboard [3rd]

🔹 AI Prediction & Comparative Analysis Dashboard [4th]

🔹 Marks & Result Analysis [5th]

🔹 AI Prediction & Comparative Analysis [6th]

---

To view the main project file: Click here

NOTE: This file requries an compatible app for opening. Refer Tech Stack for more information.

---

Project Structure

student-ai-workflow/
├─ data/
│   ├─ raw/
│   │   └─ Student_Marks_Result_Analysis_MS-ELEVATE_CO-PILOT_PROJECT_DATASET_RISHIT-GHOSH.xlsx
│   └─ processed/
│       ├─ student_predictions_20260115_151427.csv
│       └─ student_predictions_latest.csv
├─ docs/
│   └─ screenshots/          # Dashboard visuals
├─ exports/
│   ├─ AI-Prediction-and-Comparative-Analysis_MS-ELEVATE_CO-PILOT_PROJECT_RISHIT-GHOSH.pdf
│   └─ Student_Marks_Result_Analysis_MS-ELEVATE_CO-PILOT_PROJECT_PRESENTATION_RISHIT-GHOSH.pdf
├─ models/
│   └─ pass_classifier.pkl   # Trained ML model
├─ powerbi/
│   └─ Student_Marks_Result_Analysis_MS-ELEVATE_CO-PILOT_PROJECT_RISHIT-GHOSH.pbix
├─ presentation/
│   ├─ Student_Marks_Result_Analysis_MS-ELEVATE_CO-PILOT_PROJECT_PRESENTATION_RISHIT-GHOSH.pptx
│   └─ Student_Marks_Result_Analysis_MS-ELEVATE_CO-PILOT_PROJECT_PRESENTATION_RISHIT-GHOSH.pdf
├─ scripts/
│   ├─ train_model.py        # Model training script
│   └─ run_pipeline.py       # Prediction and export pipeline
├─ venv/                     # Python virtual environment
├─ .gitignore
├─ LICENSE
├─ README.md
└─ requirements.txt

---

---

How to Run

1. Set up environment

python -m venv venv
.\venv\Scripts\Activate   # On Windows

2. Install dependencies

pip install -r requirements.txt

3. Train the model

python scripts/train_model.py

4. Run prediction pipeline

python scripts/run_pipeline.py

5. Open Power BI dashboard

• Navigate to powerbi/Student_Marks_Result_Analysis_MS-ELEVATE_CO-PILOT_PROJECT_RISHIT-GHOSH.pbix
• Click Refresh to load latest predictions

---

Dashboard Pages

1. Marks & Result Analysis

• Rule-based evaluation of student marks
• KPIs: Total Students, Average Marks, Pass %, Top Scorer
• Visuals: Pass/Fail pie chart, subject averages, student table

2. AI Predictions & Comparative Analysis

• AI-driven predictions vs rule-based outcomes
• KPIs: Predicted Pass %, Average Overall Marks
• Visuals: Grade distribution, subject averages, pass % comparison
• Student-level prediction table with conditional formatting

---

How It Works

This project automates student performance analysis using a modular AI workflow built with Python and Power BI. Here's a breakdown of how the system operates:

1. Raw Data Ingestion

• The student marks dataset (.xlsx) is stored in data/raw/.
• It contains subject-wise scores for each student, along with identifiers like StudentID and Name.

2. Preprocessing & Feature Engineering

• The script run_pipeline.py reads the Excel file and computes:
  • Subject-wise average marks
  • Overall average score
  • Minimum subject mark
• These features are used to derive rule-based outcomes:
  • Rule_Pass: Pass if OverallAvg ≥ 40 and MinSubjectMark ≥ 35
  • Rule_Grade: Grade assigned using thresholds → A (≥75), B (60–74), C (<60)

3. Model Prediction

• A pre-trained Logistic Regression model (pass_classifier.pkl) is loaded from models/.
• It predicts whether a student will pass based on the engineered features.
• Predicted grades are also assigned using the same thresholds on OverallAvg.

4. Exporting Results

• The final student-level view includes:
  • Rule-based vs Predicted Pass/Fail
  • Rule-based vs Predicted Grade
• Two CSVs are saved in data/processed/:
  • student_predictions_<timestamp>.csv → For historical tracking
  • student_predictions_latest.csv → Used by Power BI for dashboard refresh

5. Dashboard Visualization

• Power BI loads the latest CSV and displays:
  • Comparative KPIs (Pass %, Grade Distribution)
  • Charts (Pie, Bar)
  • Student-level performance table with conditional formatting
• Two dashboard pages:
  • Marks & Result Analysis (Rule-based)
  • AI Predictions & Comparative Analysis (ML-based)

---

Notes

• Sample tables show 7–8 rows for layout clarity; full dataset is available in data/processed/
• Dashboard visuals are stored in docs/screenshots/
• Project authored by Rishit Ghosh as part of the Microsoft Elevate Internship

---

---

Author

Developed by RISHIT GHOSH
B.Tech CSE (AI & ML), Geethanjali College of Engineering and Technology
MS Elevate Internship | Capstone Project | December 2025 Batch | Microsoft Copilot
Check out my Portfolio here.

---

License

This project is for academic and internship purposes.
Feel free to reference or adapt with proper credit.

---

🔗 References

• Co-pilot Studio
• Microsoft Learn – Github Co-pilot

NOTE: The Dataset used for this Project was not taken from any other source.

---

---

Extras

Notes on pass_classifier.pkl

The file pass_classifier.pkl is a serialized machine learning model created during training.
It contains the logic and parameters of the classifier used to predict whether a student will pass (1) or fail (0) based on their marks.

• How it’s created:
  Generated by running train_model.py, which trains the model using scikit-learn and saves it with joblib or pickle.

• What it stores:

  • The chosen algorithm (e.g., Logistic Regression, Decision Tree, etc.)
  • Learned weights/parameters from the training dataset
  • Any preprocessing steps included in the pipeline

• How it’s used:
  When you run run_pipeline.py, the .pkl file is loaded and applied to new student data to generate predictions (Predicted_Pass, Predicted_Grade).

Example usage:

import joblib

# Load trained model
model = joblib.load('models/pass_classifier.pkl')

# Predict outcomes
predictions = model.predict(X_test)

This ensures consistency — the same trained model can be reused without retraining every time.

---