Machine Learning & Deep Learning Portfolio

Sentimental analysis Amazon review data

Technologies: Parallel Deep Learning, Python, LSTM, PyTorch, Dask, GPU cluster, Linux
GitHub Repository: Link to Repository

• Implemented scalable sentiment analysis pipeline using PyTorch and distributed computing (Distributed Data Parallel and Model Parallelism) to achieve 95.17% accuracy
• Optimized preprocessing and training workflows by leveraging Dask for distributed data processing and mixed precision training, reducing training time by over 50% on multi-CPU and GPU setups

Transformer Chatbot

Technologies: Deep Learning, Python, Transformers, PyTorch, GPU cluster
GitHub Repository: Link to Repository

• Built a Transformer-based chatbot model in PyTorch, leveraging tokenization, positional encoding, and self-attention
• Achieved 99% training accuracy over 300 epochs by optimizing model with cross-entropy loss and Adam optimizer

Text Classification using BERT (Financial Data)

Technologies: LLM models, HuggingFace BERT, Resampling, PyTorch, Scikit-Learn, Machine Learning
GitHub Repository: Link to Repository

• Achieved a 95% accuracy rate by fine-tuning Transformer-based LLM models (BERT, DistilBERT) for text classification of financial data into predefined classes (Noise, Text, Table)
• Enhanced model performance by implementing k-fold cross-validation for robustness and applying resampling techniques to address class imbalance in the dataset

British Airways Review Analysis

Technologies: Sentiment analysis, Data Scraping, Data visualization, Transformer model, Machine learning, SciPy
GitHub Repository: Link to Repository

• Created an interactive dashboard analyzing over 150,000 electric vehicles, providing insights into market trends and technological advancements
• Enhanced decision-making by visualizing key metrics, including over 200% increase in BEV adoption and state-wise vehicle distribution from 2020, using dynamic charts and maps