Vijay's Coding Portfolio


Table of Contents

Name What is it? Link
Cognateful LLM-Powered Language Learning View
NeRFs from Scratch Implemented NeRFs (Neural Radiance Fields) to reconstruct a rotating 3D scene from 2D images. View
Flow Matching Models from Scratch Implemented a flow-matching generative model with classifier-free guidance (CFG) to synthesize novel handwritten digits from noise. View
GSO: Challenging Software Optimization Tasks for Evaluating SWE-Agents Designed automated pipeline to evaluate SWE agents View
PintOS (Operating Systems Class Project) Implemented kernel extensions for the PintOS operating system View
RISC-V CPU Designed pipelined RISC-V CPU in Logisim View
FactGrid Cuneiform Project ML model to segment & transcribe Cuneiform tablets View
The David Deutsch Dictionary Online database of quotes with Google Sheets backend View


More Info

Project Name
Image
Description
Image from my language learning app
Interactive language-learning app that teaches you French using stories written in French, with translation exercises that are scored using large language model judgements.
A gif of a spinning lego
Designed and implemented a NeRF (Neural Radiance Field) model in PyTorch based on the original NeRF paper. Successfully reconstructed a 3D scene of a rotating LEGO truck from a set of 2D views.
An image being generated from scratch
Implemented flow matching generative model in PyTorch using a time and class-conditioned UNet. Trained the model on MNIST with conditional dropout to support classifier-free guidance during inference. Built sampling + visualization pipeline to generate GIFs showing emergence of digits over time.
An image of the GSO task generation pipeline
Paper Accepted at NeurIPS 2025 (Datasets & Benchmarks Track): Coauthored Global Software Optimization (GSO), an LLM code optimization benchmark spanning 102 optimization problems and five programming languages.
PintOS (Operating Systems Class Project)
An image of the PintOS operating system
Built a transaction-safe Unix-like file system with hierarchical inode structure, resizable files, and rollback recovery. Developed an in-memory buffer cache with fine-grained locking for concurrent disk I/O. Implemented a POSIX-style threading subsystem with kernel-level synchronization primitives and priority scheduling to enable true parallel execution.
RISC-V CPU
RISC-V CPU Datapath
Built a full CPU in Logism that runs instructions in assembly language. Implemented two-stage pipelining to increase throughput. Prevent control hazards via pipeline flushing.
FactGrid Cuneiform Project
Develop machine learning model to segment and transcribe dataset of 1,068 annotated Cuneiform tablets using the YOLOv8 architecture.
The David Deutsch Lexicon
Online database of quotes from science author and quantum physicist David Deutsch. Utilizes Google Sheets as a backend database to allow for seamless and near-real-time updates. Scraped quotes off the internet using Python3 and NLTK (Natural Language Tool Kit). Designed UI using Flask and HTML.