Mondol Mridul Provakar

I am a passionate Researcher and Developer focused on Advancing Technology through Artificial Intelligence and Machine Learning. I hold an Undergraduate Degree in Computer Science and Engineering from Rajshahi University of Engineering & Technology (RUET), Bangladesh. I've published Two Conference Papers on the Effectiveness of Large Language Models in Financial Question Answering and Text Summarization. I have implemented projects across diverse fields including Natural Language Processing, Time Series Analysis, Computer Vision, Machine Learning, CPU Design, Android and Web Development. I have a deep passion for music and I enjoy traveling, playing video games and reading books.

Research Interests

• Large Language Models
• Natural Language Processing
• Recommendation Systems
• Autonomous Cybersecurity
• Reinforcement Learning
• Computer Vision
• Human Computer Interaction

Experience

Publications

Projects

Preprocessing, Fine-Tuning, and Conversational Retrieval with Llama-2

2024

• Utilized Transformers for NLP tasks including sentiment analysis, zero-shot classification, text generation, mask filling, named entity recognition, question-answering, summarization and translation with pre-trained models.
• Built a LangChain system with the Bloom model using prompt templates to generate multi-question responses.
• Fine-tuned BERT on SQuAD achieving 81.18% Exact Match and 88.67% F1, highlighting robust question-answering performance.

GitHub

Personalized Item Recommendation Using T5

2024

• Built a recommendation system using T5 with sequence-to-sequence fine-tuning.
• Implemented functions to detect unpurchased items, structured prompts, and tokenized data with truncation and padding.
• Trained with Seq2SeqTrainer to generate recommendations, compiling predictions into a DataFrame for evaluation.

GitHub

Foundations of Neural Network Algorithms from Scratch

Labwork | 2023

• Developed K-Nearest Neighbor, Single Layer Perceptron, Multi-layer Neural Networks, Kohonen Self-organizing Networks, and Hopfield Networks from scratch.
• Evaluated models using the Breast Cancer dataset.

GitHub

Information System Analysis and Design: A Comprehensive Study of Rajshahi Krishi Unnayan Bank (RAKUB)

Group Assessment: Labiba Mashiat Mahi, Mondol Mridul Provakar, Soumit Das, Chinmoy Mitra | 2023

• Conducted a comprehensive analysis to enhance RAKUB’s (Rajshahi Krishi Unnayan Bank) operational efficiency.
• Covered feasibility, data flow, cost-benefit analysis, form design, and system development.
• Delivered recommendations for improved banking services to support Bangladesh’s agricultural sector.

GitHub

Kidney Disease Classification using CNN & VGG16 CNN

2023

• Used VGG16 CNN and a custom CNN to classify kidney conditions (Normal, Cyst, Tumor, Stone) with high accuracy.
• Preprocessed images by rescaling, resizing, and one-hot encoding with an 80/10/10 split for training, validation, and testing.
• Achieved top accuracy with CNN at 99.92%, while VGG16 scored 99.91%, both models demonstrating high precision and F1 scores.

GitHub

JFK Airport Visibility Prediction Using Time Series Forecasting

Coursework Project | 2023

• LSTM and GRU models were developed for time series forecasting, with both univariate and multivariate approaches to predict hourly visibility at JFK Airport.
• Cleaned, indexed, and visualized the dataset; applied a sliding window approach to create input sequences.
• In multivariate analysis, GRU outperformed LSTM with a lower test RMSE (GRU: 1.1390, LSTM: 1.1556).

GitHub

6-Bit CPU Design (Computer Architecture)

Individual Assignment | 2023

• Designed a 6-bit CPU architecture focusing on core components such as the Arithmetic Logic Unit (ALU) and registers to perform basic operations like addition and logical operations.
• Implemented a simple instruction set architecture (ISA) supporting various modes, enhancing the CPU's capability to execute instructions efficiently.
• Utilized Digital design software to create and visualize the CPU circuit, providing practical experience in computer architecture and digital system design.

GitHub

SigPlanet: Real-Time Sign Language Detection

Coursework Project | 2022

• Built a real-time gesture and letter recognition system using Mediapipe and LSTM, processing video input for live predictions.
• Utilized Mediapipe to extract 3D hand landmarks storing keypoints as numpy arrays with labels for training.
• Designed a multi-layer LSTM model with categorical cross-entropy loss and Adam optimizer, achieving high-accuracy predictions displayed on screen.
• Integrated with a Django-based website using HTML and CSS for accessible, interactive gesture recognition.

GitHub

Soil Strength Prediction using Machine Learning Regression Models

2022

• Conducted regression analysis on the "F_CLAY_7_216_TC304" dataset to predict soil test results (su(test) in kPa).
• Managed missing values, eliminated unnecessary features, and standardized the dataset.
• Evaluated multiple regression models, with Support Vector Regression achieving the best results: RMSE of 4.402 and R² of 0.754.

GitHub

Comprehensive Machine Learning Algorithms & Practical Projects

2022

• Algorithms Implemented: Gradient_Descent | Linear_Regression | Logistic_Regression | Naive Bayes | KNN | Decision Tree
• Projects Done: Titanic - Machine Learning from Disaster | Mercury prediction | Sentiment Analysis with Naive Bayes

GitHub

Smart Student Guide: Enhancing University Connectivity for Students

Coursework Project | Group: Mondol Mridul Provakar, Mahbuba Sultana Prangan | 2021

• Developed an Android application to enhance student-university engagement through features like notice boards, video lectures, eBooks, and faculty directories.
• Implemented separate dashboards for admins to manage notices and updates, and for students to access campus information, maps, and resources.
• Created intuitive navigation with a home screen, drawer menu, and organized sections for easy access to university resources and announcements.

GitHub

Education

Certification

Honors and Awards

Extracurricular activities

References

Contact

• Phone: +880 1941715784
• Email: mondolmridul007@gmail.com
• LinkedIn: linkedin.com/in/mondolmridul007
• Address: 76 Palpara Noapara South, Abhaynagar, Noapara-7460, Jashore, Khulna, Bangladesh