My Projects

Built an interactive, explainable AI dashboard for real-time credit risk prediction using XGBoost, Random Forest, and Logistic Regression, with full model evaluation and comparison.

Designed and implemented a lightweight ML pipeline orchestrator entirely from scratch to automate and manage end-to-end ML workflows without external orchestration frameworks.

Engineered a real-time ML pipeline with MLflow for experiment tracking, version control, and continuous integration, inspired by GitOps practices for production-ready deployment.

Built a full-stack ML inference deployment pipeline supporting real-time sentiment analysis using FastAPI.

Built a RAG-based chatbot leveraging LangChain and the Mistral-7B model to generate accurate, context-aware medical responses by combining LLMs with vector search databases.

Built an ML inference pipeline capable of operating directly on encrypted data using the CKKS scheme from the PySEAL library, ensuring complete data privacy throughout the workflow.

Implemented a feature extraction pipeline for CIFAR-10 images by flattening pixel data, normalizing the data and applying standard augmentations for robust classification.

Designed an end-to-end machine learning pipeline for predicting NBA game outcomes using real-time and historical player behavior data, achieving 83% accuracy.

Reproduced FixMatch results on CIFAR-10 using Wide ResNet-28-2, achieving 90% of the performance reported in the original paper.

Developed a pipeline for generating stabilized panoramas from videos using modular components, reducing processing time by 25%.

Designed and implemented an end-to-end automated data pipeline using Apache Spark and Google Cloud Dataproc to efficiently process and transform large-scale datasets ingested from GCS.

Built a neural network framework from scratch, implementing core components, achieving 95% accuracy on MNIST dataset.

Led research to develop a framework for automated implicit bias annotation using fine-tuned LLAMA-2 models with QLoRA and PEFT for efficiency and scalability.

Built a fraud detection system processing financial transactions using supervised and unsupervised learning.

Performed exploratory data analysis and feature selection on the NSL-KDD dataset, improving feature relevance and reducing dimensionality by 15%.

Conducted a study on advanced image captioning techniques using RNNs, BiLSTMs, and Transformers, achieving a 15% improvement in BLEU score over baseline RNN models.

Developed a salary prediction tool using Glassdoor data to estimate salaries for data science roles.

Surveyed standard approaches for monocular depth estimation, leading to a published survey paper analyzing architectural trade-offs.

Developed an automated data pipeline to integrate data from AWS S3 and Azure Blob Storage into a centralized GCS Bucket staging area.