Here are some of the things I’ve built — usually from scratch, just the way I like it

---

Deep Learning

Explorations in frameworks, NLP, and training workflows.

PureTorch

A PyTorch-like deep learning framework implemented from the ground up.
Focuses on building autograd, tensor ops, and training loops with minimal dependencies.

Inqueropedia

A Wikipedia-powered chatbot that uses custom tokenization and search — no pre-trained models or black boxes.
Still in the works — currently figuring out a way to train the model :(

---

Deployment Demos

Live demos hosted on Hugging Face Spaces.

Food Classifier

An EfficientNet-B2 model to classify between 3 food types: pizza, steak, sushi.

Food Classifier 101

Trained on a 20% subset of the Food101 dataset.
Achieved a 1.59% improvement in top-1 accuracy over the original paper’s baseline.

---

Deep Learning Paper Replications

Replicating deep-learning papers from scratch.

Vision Transformer (ViT)

A faithful implementation of the ViT paper, built layer-by-layer with annotated code and experiments.

Attention Is All You Need (Transformer)

Recreated the original Transformer architecture from scratch, focusing on attention mechanisms and positional encoding. Link to paper

Deep Residual Learning for Image Recognition (ResNet)

Replicating the work from the ResNet paper on the tiny-imagenet dataset. Still in progress — model training and evaluation ongoing.

---

Hardware & RC Cars

3D-Printed RC Builds

Designed and built multiple RC cars using Fusion 360, custom chassis, and open-source electronics.
Some builds use salvaged parts, others are entirely custom.

---

Side Quests

Planner App (JavaFX)

A small productivity tool written in Java, built to manage tasks and deadlines with a clean desktop UI.