Hello, I am Ved Chugh

1 of 5: Smart Lock

Last year, my apartment building charged me one hundred dollars for a spare key. Instead, I built an end-to-end facial recognition smart lock integrating computer vision, software engineering, and custom hardware. The system verifies identity using facial embeddings, includes personally-created blink detection to prevent spoofing, and physically unlocks a door using a microcontroller and a 3D-printed servo mechanism. I also built a lightweight frontend with a live MJPEG video stream for real-time monitoring and notifications. It took months, requiring tons of patience, but in the end I was rewarded with a shiny new lock and new insight into computer systems. This is the project that makes me feel the most pride.

2 of 5: Data Science Crash Analysis

Engineered a data analysis pipeline using Python (Pandas, SQL) and machine learning (scikit-learn’s Random Forest) to identify predictors of U.S. car crashes. Delivered actionable insights via interactive visualizations that simplified complex trends. Collaborated using Git for seamless version control, ensuring reproducibility. Our results underscored environmental factors’ impact on collision rates. Explore the project’s code by clicking on the image.

3 of 5: Computer Vision Movie Analysis

In college, filmmaking enamored me. Since, graduating I've become obsessed with merging my degree in computer science with my studies in filmmaking. To start, I designed and implemented a neural netowrk from scratch using PyTorch to identiy movie eras based solely on poster visuals. Later, I implemnted transfer learning wth a pretrained ResNet, fine tuning layers for era-specific pattern recognition. However, data exploration was the focus of this project, and allowed me to get better insight into the workings of a digital image. In the end, this project serves as a major stepping stone towards my goal of computer vison expertise.

4 of 5: Machine Learning Intro

Implemented a neural network for classification tasks, like recognizing handwritten digit images. Computed gradients via backpropagation using vectorized NumPy operations, handled batched inputs, and updated parameters through gradient descent. The modular design allows swapping activation functions and loss functions on the fly. It highlights my foundational knowledge of neural network mechanics and software design for machine learning systems.

5 of 5: Frontend Training

A diverse portfolio of frontend projects showcasing technical growth and modern tool proficiency built using vanilla JavaScript, React, CSS, and Next.js. Concepts include component-based architecture, responsive design, and CMS integration. Continuously expanding my toolkit to align with industry demands, I emphasize clean code and user-centric solutions using industry tools. This portfolio is actively evolving—click the image to explore real-time updates on my growth.