Selected builds
WORK // PROJECTS
Sub-Millisecond Voice Activity Detection (VAD) with Custom Wakeword
A lightweight, edge-optimized Voice Activity Detection and custom wakeword recognition engine built from scratch in PyTorch. Features a Broadcasted Residual Network (BC-ResNet) capable of sub-millisecond inference and utilizes custom Focal Loss to handle massive dataset imbalances.
Artificial Neural Networks from Scratch
Engineered a complete Artificial Neural Network from scratch using pure mathematics and NumPy. Implemented core components including Dense Layers, Backpropagation, Binary Cross Entropy Loss, Softmax activation, and the ADAM optimizer without relying on high-level deep learning frameworks.
Real-Time AI Calling System
Architected a low-latency AI calling system enabling seamless voice interactions. Developed during my AI Internship at SiyaraTech Innovations. (Note: As this is a proprietary company project, the source code is not publicly available).
Full-Stack RAG Chatbot Engine
Architected an asynchronous, multi-modal ingestion pipeline that processes complex PDFs and generates accurate AI responses backed by precise, verifiable document citations.
RAG-Powered WhatsApp E-Commerce Bot
Built a generative AI chatbot for WhatsApp to automate customer queries and facilitate e-commerce flows. Developed during my AI Internship at SiyaraTech Innovations. (Note: As this is a proprietary company project, the source code is not publicly available).
Hibiscus CNN Classifier
A deep learning-based Convolutional Neural Network (CNN) model for classifying Hibiscus leaf species (Rosa-Senis, Canabinus, Mutabilis) using image data. Trained on 5,100 samples using morphological features like petal length, color, and sepal length.
ANN for MNIST Dataset from Scratch
Built a custom Artificial Neural Network from the ground up using pure NumPy to classify the MNIST handwritten digit dataset, implementing mathematical foundations for forward passes, backpropagation, and optimization.
ClassMantra AI Educator Assistant
Developed a full-stack educational web application featuring automated assignment grading and AI-generated feedback mechanisms.
Zero: The Focus App
A productivity and focus application featuring a minimalist UI design heavily inspired by the Nothing OS dot-matrix aesthetic.