A lightweight feed-forward neural network implemented from scratch in C++ and CUDA. Includes GPU-accelerated linear layers with cuRAND-based weight initialization and manual memory management. Designed for learning and demonstrating low-level deep learning internals.
This is a simple and basic implementation of a ChatBot using Transformer LLM and a Cornell Movies Dataset.
Raspberry Pi 5 autonomous car using the Raspberry Pi AI Kit (Hailo-8L) to run a Keras NVIDIA behavioral-cloning model compiled to .hef. Frames are captured with Picamera2, preprocessed (crop→YUV→blur→resize), inferred in real time via HailoRT’s InferModel API, and the predicted steering angle is mapped to L298N dual-motor PWM (0–255)
This notebook demonstrates how to fine-tune the Phi-3.5-Mini-Instruct model using LoRA (Low-Rank Adaptation) for a domain-specific task: creating a lightweight medical assistant.
A basic OS that boots from nothing and features memory allocation, keyboard and time interrupts and shell.
This notebook contains a from-scratch reimplementation of the paper: An Image is Worth 16×16 Words: Transformers for Image Recognition at Scale (Dosovitskiy et al., 2020).