Rasya Akbar
Lazuardi

About - p.01

Where I work from

I'm a university student and full-stack developer, most at home in React and Next.js, building interfaces that are fast and simple to use. Over the past year I've also picked up Python, PyTorch, and discrete-event simulation, working on a research project that models how software teams actually plan their sprints. I like taking a system apart to understand it, then putting it back together as something people can actually use.

- currently splitting weeks between shipping features and running simulations

Projects - p.02

Things I built

Mostly built to answer "could I make this myself".

TechNotes

a sticky-note board that syncs

A better replacement for the sticky notes stuck to everyone's monitor, login with JWT cookies, and every update syncs live across users.

MongoDB · Express · React · Node · JWT

TechNote-API ↗ TechNote-front-end ↗

Discord Twin

rebuilding Discord to understand it

Real-time chat with Socket.IO and API-driven video calls, the fastest way I found to actually understand how chat apps stay in sync.

Next.js 13 · Prisma · MySQL · Socket.IO

discord-twin ↗

Web Builder

drag, drop, export, repeat

A static site builder with live preview and a downloadable export, built to make putting a simple website together painless, with a paid tier through Midtrans.

Next.js 14 · Prisma · MySQL · Midtrans

Web-Builder ↗

Farm-to-Table Marketplace

cutting out the middleman

A prototype marketplace that lets farmers list and sell produce straight to customers, no distributor in between. Built as a proof of concept - not yet in production.

PyTorch Learning Log

Learning

teaching myself transfer learning

A set of Jupyter notebooks working through PyTorch fundamentals - building small deep learning models from scratch, then transfer learning, which turned out to matter a lot more once I started applying models to real data.

Python · PyTorch · Jupyter

learning-pytorch-deeplearning ↗

Research - p.03

PuTI DEVTI - Sprint Decision Intelligence Agent

Working alongside a faculty advisor, I'm building a digital twin of a real product team's sprint workflow - simulating how work actually moves through backlog, in-progress, and done using discrete-event simulation, run across Monte Carlo replications for six different sprint scenarios. On top of the twin sits a two-tier recommendation engine: a Random Forest model backed by a heuristic fallback for when the data runs thin, so it can suggest sprint decisions without pretending to be more certain than it is.

I wrote up the full reasoning behind every modeling choice in a 26-page walkthrough, and built forward-looking logging templates so the team can start feeding it real sprint data going forward.

- the honest part: the model says "I don't know" when it should, instead of guessing.

Field notes - p.04

Volunteer & teaching

Both through community projects led by a professor I work with - plus some mentoring of my own.

LeetCode - p.05

Grinding, honestly

Easy is easy, medium I can mostly close out, hard is where I actually learn something.

Contact - p.06

Send a note