Nethum Weerasinghe

Developer, Texas A&M

Studying computer science at Texas A&M and building tools for the way software actually gets written now, moving between systems, interfaces, and the agents that sit between them

Right now I am building Seer, a shared window onto the terminals running on one server, alongside Peanut, a review surface where people and coding agents work the same document

Away from my own projects I contribute to the GPU libraries at NVIDIA and RAPIDS, and I have spent time on production systems in industry and on medical imaging research in a university lab

I share most of what I build on GitHub and can be reached by email

Resume

Contributions

Projects

SeerNewA shared window onto every terminal running on one server, where the only permission you hand out is the right to type

PeanutA multiplayer review surface where people pin instructions to a live document and a coding agent applies them in rounds

StampedA multi-role onboarding platform for financial institutions and their vendors, and first place in the Goldman Sachs track at HackUTD

VeroA drug interaction checker that runs Monte Carlo simulation over a personal risk profile instead of returning one more yes or no

MerchantAn iOS hub for a wallet full of credit cards that says which one to swipe before you reach the counter

Open source

NVIDIA/cudfGPU DataFrame library · 9.7k stars6 mergedClosed a pandas compatibility gap that was blocking GPU migration, and removed a required disk round-trip by letting an ingestion API take byte buffers directly

rapidsai/rmmRAPIDS Memory Manager · C++ and CUDA3 mergedLet a size-capped GPU memory pool reach its configured maximum by reclaiming freed blocks before rejecting a request, and enforced compile-time detection of discarded allocations across the public API

NVIDIA/ccclCUDA Core Compute Libraries · 2.5k stars1 mergedCut repeated compilation overhead by caching compiled kernels, replacing a rebuild on every call with a cache lookup

NVIDIA/cumlGPU machine learning · 5.2k stars1 mergedCorrected a naive Bayes defect that returned silently wrong predictions by preserving established class labels across incremental training batches

Experience

Elysian

Design engineer, contract

2026-Now

Designed and built the storefront for an ecommerce label running at $500K ARR with more than 50,000 visitors a month
Shipped the whole shopping surface in static HTML, CSS, and JavaScript ahead of the move to Shopify Liquid, covering product pages, cart, search, and the phone layout

Global Shop Solutions

Software engineering intern

May-Aug 2026

Co-built and launched Traveler, a Next.js project management platform that replaced Monday.com for around 20 users and saved $25K a year
Built a token-efficient MCP server so the same team could manage tickets through AI agents, with over 90% of updates going through it
Refactored data access managers in customer-facing production code, cutting code volume by over 60% and data retrieval time by 34%

Sketch Recognition Lab

Undergraduate researcher

Jan-May 2026

Engineered an end-to-end machine learning pipeline for medical image segmentation, from preprocessing and embedding caching through training, Dice/IoU/HD95 evaluation, and visualization
Ran training and evaluation on the Grace HPC cluster at Texas A&M with A100 GPU jobs and Slurm