README.md

metadata

title: Smart MCQ Solver
emoji: 🧠
colorFrom: blue
colorTo: purple
sdk: gradio
app_file: app.py
pinned: false

Smart MCQ Solver (BiLSTM)

This is the final deep learning project for the DL & GenAI course.

Deep Learning & Generative AI Project

Student Information

Name: Varnit Chourasiya Roll Number: 23f3000843 Course: Deep Learning & Generative AI Project (BS in Data Science and Applications)

---

Project Overview

This project focuses on developing and evaluating multiple Deep Learning and Generative AI approaches for a Natural Language Processing (NLP) task. The workflow includes data preprocessing, semantic similarity analysis, transformer-based models, retrieval-augmented generation (RAG), model fine-tuning, and ensemble techniques.

The project is part of the Deep Learning & Generative AI curriculum and will be evaluated through Kaggle performance, GitHub repository quality, experiment tracking, report submission, and viva examinations.

---

Objectives

• Perform text preprocessing and feature engineering.
• Build baseline NLP models using traditional embedding techniques.
• Explore transformer-based architectures such as BERT and RoBERTa.
• Implement Retrieval-Augmented Generation (RAG) pipelines.
• Fine-tune pretrained language models.
• Compare multiple models using standard evaluation metrics.
• Track experiments using Weights & Biases (W&B).
• Improve prediction performance through ensemble methods.

---

Technologies Used

• Python
• Pandas
• NumPy
• Scikit-learn
• PyTorch
• Hugging Face Transformers
• Weights & Biases (W&B)
• Kaggle
• Git & GitHub

---

Repository Structure

.
├── data/
│   ├── raw/
│   └── processed/
│
├── notebooks/
│
├── src/
│
├── models/
│
├── reports/
│
├── screenshots/
│
├── requirements.txt
│
└── README.md

---

Project Milestones

• Milestone 1: NLP Foundations & Semantic Similarity
• Milestone 2: Transformer-Based Models
• Milestone 3: Retrieval-Augmented Generation (RAG)
• Milestone 4: Model Fine-Tuning
• Milestone 5: Ensemble Learning
• Final Submission & Evaluation

---

Experiment Tracking

All experiments, training runs, and model comparisons will be tracked using Weights & Biases (W&B).

---

Kaggle Competition

The final model performance will be evaluated through the official Kaggle competition associated with this project.