Xander's Scribblings

Essays, ideas, and explorations in technology, mathematics, and creative thinking

Hi, I’m Xander van Pelt. I’m a maker, builder, and full-stack developer passionate about creating tools that solve real problems. I started coding at 13, reverse-engineering game engines and crafting small 2D Java games before diving into full-stack development, AI integration, and automation systems.

I’ve built projects like ClassroomFeed, which automates Google Classroom data into AI-driven weekly summaries; the MekongMemo pipeline, a multi-agent LLM system that aggregates and synthesizes regional news; and TutorBoard, an app used by teachers and students to streamline class information; and more to come… My projects have taught me how to integrate Google APIs, design relational databases, deploy scalable backends, and weave large language models into functional systems safely and effectively.

Alongside web systems, I’ve explored optimization research, signal processing, and graphics programming, driven by curiosity about how math, code, and art converge. I’m currently pursuing IB Diploma HLs in Chemistry, Physics, and Math AI, and plan to study Mechanical or Computer Engineering in university, where I can merge my passion for intelligent systems, automation, and design.

Next, I’m working on creating a morphological latent space representing to model Jansen linkage parameters and foot trajectory similarities.

Read my recent post on creating systems that utilize LLMs here or my exploration into Fractal PCA based image compression here.

Projects and Creations

ClassroomFeed.com

image

After getting tired of scrolling through 10 separate Google Classroom feeds and forgetting about deadlines, I decided to automate with a Google Classroom API–integrated app that uses AI, your inbox, and some magic to generate weekly progress reports, reflections, and recaps on what work you’ve missed, completed, or is soon to come. AI tracks reflections and progress over the weeks using the data you provide it.

This gave me the skills to:

With a growing user base in over 10 schools across the world, this SaaS project has been a culmination of my technical skills applied in a real-world context, providing an actual product with appeal.

ClassroomFeed.com
(Repository is private, DM or email to request viewing)

MekongMemo Automation Pipeline

image

Developed a proprietary interface and system to aggregate, scrape, store, categorize, score, group, and synthesize stories from multi-perspective publication narratives sourced from RSS feeds across Southeast Asian countries. The system uses a Discord bot as a front/interface to operate the robot. Coded 100% in Python with MySQL for storage. Anthropic and OpenAI LLM API integrations, LLM prompting.

I worked on this project over the summer at 16–17 years old, where I developed a pipeline for creating and synthesizing multiple multi-perspective articles from varied news sources into a single newsletter. This project was completed at a time when “vibe-coding” and using AI to write the majority of code (beyond just debugging) was not a thing. I’m taking full credit for writing every single line of this one.

This project taught me:

Mekong Memo Newsletter
(Repository is private, DM or email to request viewing)

Neutron Java Game Engine

image

One of my first large-scale passion projects. I began this endeavor at 14 years old. After competing in a Game Jam, I fiddled with Unity and watched a tutorial but was fed up with the black-boxed, abstract tool I was wrestling with. I wanted to work with my bread and butter: raw code. Though others had submitted entries full of 3D models and fancy shaders, my game was crude; it wasn’t good and won me nothing. But in retrospect, that game—a bouncing blue square among red enemies—made me realize my true passion. It wasn’t about making a pretty or super fun game, but about all the implementation nitty-gritty along the way. Considering things like whether a projectile should ask an enemy “are you dead?” or whether the enemy should broadcast “I’m dead” to anyone listening—that was the real game I wanted to play.

I programmed pixel shader code, collision checks, and game loops by reading articles, Stack Exchange posts, watching YouTube videos, and learning Object-Oriented Java. I find game engines to be the most beautiful expression of Object-Oriented Programming. I have a fully working Flappy Bird demo, with collisions, audio, UI, etc., up on my repo.

This game taught me the absolute core and bread and butter of object-oriented programming. I look back in nostalgia at the inspiration and passion this project brought. It became a black hole of implementation specifics and system-level considerations. Every system to implement was a delicious puzzle, truly. I love this project.

I learned:

Repository, Website

Research Paper Comparing Optimization Methods for Jansen’s Linkage

final_algorithm_comparison image

Optimizing Jansen’s linkage proportions to achieve a desired foot trajectory using a gradient descent and genetic-based approach and comparing effectiveness, measured in final RMSE between the desired path and actual.

This project began as an applied problem for comparing GD and GA for my IB Extended Essay (it was overkill), but after completing it, I thought I should publish it as a full research project because the idea was novel (though by no means cutting-edge).

I created numerical simulations to model Jansen’s Linkage, implemented gradient descent using approximated finite-difference gradient methods, and used PyGAD for the genetic implementation. I learned the essentials of optimization, multivariable differential calculus, and extended my personal learning into neural network training I had done.

This project taught me:

The full paper is available here: (Optimizing Jansen’s Linkage: A comparison of Gradient Descent and Genetic Algorithms)[https://doi.org/10.5281/zenodo.18118038] Code is here: github.com/XenenDev/Jansen-Linkage-Research

HHIS TutorBoard

image

I started this project at the beginning of my IB course. Frustrated with Google Classroom’s shortcomings and the disorganized notification and announcement system in my class (a maze of emails, notices, and word of mouth), I wanted to create a solution that brought all these information sources into one coherent system, designed for morning tutors to read through with the class. I applied the skills I learned from building the MekongMemo pipeline to this project.

The skills I learned here allowed me to apply them to ClassroomFeed.com. Teachers and students loved the app—they used it to check air quality, filter announcements for their class specifically, and circumvent the scattered and cluttered previous system.

Website
(Repository is private, DM or email to request viewing) Full article on what I learned about building AI systems from this

PCA Based Transform With Fractal Patches for Image Compression

atlas_comparison pca_vs_dct_visual_comparison

After being obsessed with DCT compression (the math behind JPEG image compression) over summer, I began experimenting with what I could do by modifying the basis vectors. I had the idea to replace the basis vectors with ones that were “learned” to capture more information with fewer basis vectors (to maximize energy compaction on a per-image basis). After experimentation and research, I realized PCA would yield the exact basis vectors I required.

The final compression didn’t perform noticeably better than JPEG and required storing the basis vectors in the image header, which cost precious bytes. Quantizing the basis vectors was also an option, but that led to a trade-off between quality loss and space saved from storing the basis vectors as integers. Instead, I explored a tangential idea using a fractal-based system to subdivide the usual 8×8 patches used in JPEG further if an algorithm found that section had sufficient detail to require subdivision. The final result was interesting, but the borders between subdivided parts showed an obvious difference in quality, making the result look blocky.

Though the project wasn’t a success, I still learned:

Full article I wrote on this

Art and Creation in Blender and Unreal Engine

render_brick image image image

I have always had an interest in 3D modelling and scene creation as a form of creating art. I have made some product renders for friends and family, shown above. In Unreal Engine, I have created a Middle Eastern desert scene, with interiors and lighting. I know the basics of lighting, 3D modelling, and how to write shaders and use these tools. I learned:

Java Sand Simulator and Cellular Automata

image

Created a Cellular Automata-based rule set to create a sandbox simulation. Modeled water, sand, and rock. Per-pixel-level handling of physics.
Taught me:

Repository

Modern Take on Rosenblatt’s Perceptron

image

After learning about how the most basic neural models work, and inspired by the Perceptron created by Frank Rosenblatt in 1957 with manual wiring and actual photosensors, I wanted to try to build a binary classifier, but with modern Python. With no data structures or Python libraries for machine learning of any kind, I tried to build it from the ground up because I wanted to understand at a fundamental level how it worked.

From this project I learned:

Here’s the result: Repository

Lazy MySQL Wizard

image

Too lazy to learn MySQL and finding myself prompting AI on how to execute certain actions on a database—always having to re-enter my database’s schema—I decided to build essentially a ChatGPT wrapper with a MySQL table viewer and interface. Now making changes to your database is as easy as a prompt and as transparent as a neat table.

Taught me:

Repository

Work Experience

Vimarn Suriya Co. Ltd.

I engineered and computationally solved the problem of allocating the most optimal and favorable parking spaces to residents of the new 69-story Dusit Residential tower.

Private real estate invoice and email automation software

Developed an invoice automation sending system for local real estate business. —

Skills Profile – Xander van Pelt

Full-Stack Development

AI & Machine Learning Integration

Google Ecosystem & OAuth

Web Technologies & Automation

Research & Optimization

Signal Processing & Computer Graphics

Game Engine Development

Programming Languages

Software & Tools

System Architecture & Integration

Problem-Solving Domains

Additional Competencies

Random Mentions

image

I made my first feature in my school’s (now defunct) student publication, Bangkok Patana Cypher Magazine. My article remains partly visible (but shortened) in this newsletter on page 13. I made a few other articles too (which are now unfortunately lost in the sands of time).

I made my first GitHub commit at 13, working on the now-defunct Hiven (a Discord clone) API, contributing a couple of getId() and getProfileURL() methods in Python.