Local RAG pipeline using Ollama, VectorDB, and reranking

Tech Stack:OllamaRAGVectorDBRerankerUV

Highlights:

• vector search with optional reranking (FlagEmbedding)
• Interactive CLI with modes: none | rerank | both
• Local models with Ollama

End-to-end computer vision pipeline for real-time solving

Tech Stack:PyTorchOpenCVCNNImage Processing

Highlights:

• End-to-end pipeline solving Sudoku puzzles
• Lightweight CNN for digit recognition and grid detection
• Computer vision techniques for image preprocessing

Three GAN architectures implemented following original papers

Tech Stack:PyTorchLightningHydraComputer VisionGAN Architectures

Highlights:

• Three GAN architectures from scratch using pure PyTorch
• Vanilla GAN, DCGAN, and Conditional GAN following original papers
• Multiple datasets (MNIST, CIFAR-10, Fashion-MNIST)

Clean implementation of "Attention is All You Need" paper

Tech Stack:PyTorchTransformer ArchitectureMulti-Head Attention

Highlights:

• Clean PyTorch implementation following "Attention is All You Need" paper
• Deep understanding of modern NLP architectures
• Complete transformer mechanisms from scratch

Comparative analysis of modern vs traditional embeddings

Tech Stack:BERTGloVePyTorchTransformersCORD-19

Highlights:

• Comparative semantic search system analyzing academic papers
• BERT and GloVe models for semantic similarity on CORD-19 dataset
• Advanced NLP techniques for information retrieval

Advanced text summarization with T5 transformer

Tech Stack:T5TransformersROUGE MetricsGradio

Highlights:

• Text summarization models trained on CNN/DailyMail and XSum datasets
• Fine-tuned T5 model with ROUGE evaluation metrics
• Interactive Gradio interface for easy demonstration

Advanced sentiment analysis platform powered by transformer models

Tech Stack:TransformersPyTorchFastAPIReactTypeScript

Highlights:

• Full-stack sentiment analysis platform for YouTube comments
• Modern React frontend with TypeScript for enhanced UX
• Scalable FastAPI backend with PostgreSQL database