Built an ATS with entity extraction and job-match scoring to automate candidate screening and improve recruiter throughput.

Modelled CDC chronic disease records to predict state-level arthritis trends and drive an early-alert dashboard for public-health planning.

Built and explored YOLO-style object detection concepts from a learning-first perspective, breaking down the detection pipeline into understandable components such as feature extraction, prediction heads, bounding-box logic, and training/inference flow. Designed as a strong foundation for production detection systems later used in real-time CV work.

Implemented and experimented with Transformer-based sequence modeling components to deepen understanding of modern NLP architectures. Explored attention-driven representation learning and token-level sequence processing as a bridge toward LLM and generative AI engineering work.

Built a human segmentation pipeline using a U-Net model trained on the COCO dataset. Implemented data preprocessing, mask generation, training/evaluation loops, and visual inspection of predicted masks to validate pixel-level performance. This work strengthens the foundation for virtual try-on, scene understanding, and production CV segmentation workflows.

Explored recurrent neural networks for sequence and temporal learning tasks, focusing on how hidden states encode context over time. Helped build a solid understanding of sequence modeling concepts later applied in LSTM-based projects and time-dependent prediction workflows.

Studied foundational GAN research papers and implemented generative adversarial networks from scratch to understand adversarial training dynamics, generator/discriminator interplay, loss functions, and stabilization techniques. Built and trained custom GAN architectures to visualize how learned representations generate realistic samples and evaluated model behavior through iterative experimentation.

Built notebook-based classification experiments to practice model training, evaluation, and iterative improvement across core deep-learning workflows. This project collection demonstrates hands-on experimentation discipline used later in production-focused ML and CV systems.

Fine-tuned a 7B LLaMA 2 model on custom conversational data to improve contextual task handling and reduce manual support for routine queries.

Developed a CCTV security system with object tracking and face recognition, deployed as Docker microservices for reliable real-time alerts.