Home on Markus Le

VegaHed

Sat, 01 Aug 2026 00:00:00 +0930

Project Overview

VegaHed is a full-stack web application inspired by the One Piece character Vegapunk, who consumed a fruit that allows him to remember everything he has ever read. However, the more knowledge he acquires, the larger his head becomes. With his intelligence, he invented Punk Record — a system that stores his knowledge on a server and links with his brain, allowing him to keep learning without worrying about the physical limits of his head.

Wumpus World Agent

Wed, 29 Jul 2026 00:00:00 +0930

Zero ML dependencies

No neural network. No reinforcement learning. No training data, no reward function, not one learned parameter. The agent imports argparse and collections, and solves the cave by rule-based inference alone.

No neural network No RL No training data Correct on the first cave

What is mine here. This is a university assignment. The Pygame emulator, the cave generator and the scoring were provided by the course. Everything the agent does, all of the perception, inference and planning in my_agent.py, is my own work.

AI Text Detector

Fri, 05 Jun 2026 00:00:00 +0930

🏆 A group project. Our team took first prize in the Neural Network and Deep Learning course competition.

Project Overview

A binary classifier that decides whether a short piece of text was written by a person or generated by a language model. The working domain was food and restaurant reviews: short pieces, averaging around 68 words, with the occasional non-English or code-switched fragment mixed in.

Two things made it interesting. The first was compute: everything had to train on a single free-tier GPU, which ruled out throwing the biggest available model at the problem. The second was generalisation: a detector that scores well on its own validation split is easy, one that survives text from unseen generators and adversarial tricks is not.

#SkillHub.md: Agent Skill Registry

Wed, 01 Apr 2026 00:00:00 +0930

Project Overview

#SkillHub.md is a Single-Page Application (SPA) built for the 2026 Product Cadet Challenge by insightfactory.ai. It lets users upload, manage, and share markdown-based skill files for AI agents like Claude Code — the files contain instructions and metadata that guide agents in performing specific tasks.

Features

Essential

User authentication with username/password
Upload markdown skill files (private by default) with automatic frontmatter parsing (name, description, allowed-tools)
Edit and delete skill files
Tag assignment and management
Publish skills as public and browse/search/download public skills from other users

Nice-to-Have

Download and view count tracking
Version history with last-edited timestamps
GitHub-style diff viewer between versions
Clone public skills with source attribution
Analytics dashboard (activity trends, top tags)

Tech Stack

Layer	Technology
Frontend	Vite + React + Tailwind CSS
Backend	FastAPI (Python)
Database	SQLite
Infra	Docker + Docker Compose

Architecture: SPA with a dedicated REST API backend. FastAPI was chosen for strong typing, async support, and auto-generated Swagger docs. React + Vite for fast iteration; SQLite for simplicity with a storage-agnostic design ready for future migration.

MarJazz: Jazz Piano Generation

Sun, 01 Mar 2026 00:00:00 +0930

Project Overview

MarJazz is a deep learning model for autoregressive MIDI music generation, trained on solo jazz piano performances from the PiJAMA dataset. The model uses a dual-block LSTM architecture with REMI tokenisation to generate coherent ~60-second jazz piano pieces.

Architecture

Token IDs → Embedding (128-dim)
 ↓
LSTM Block 1 → Dropout → LayerNorm (pitch contour, rhythm)
 ↓
LSTM Block 2 → Dropout → LayerNorm (phrases, harmony)
 ↓
MLP Head (GELU) → Dropout → LayerNorm
 ↓
Linear → Logits (vocab_size)

Tokenisation: REMI (REvamped MIDI-derived) — encodes music as Bar, Position, Pitch, Velocity, and Duration tokens, preserving bar-level structure essential for jazz phrasing.

🦆 NexusS: Merging Study Sessions

Mon, 01 Sep 2025 14:18:10 +0930

🚀 Built in 34 Hours
🏆 BIG Hackathon Project
✨ Theme: Merger

“Merge your study. Meet your people.”

🌟 Project Overview

NexusS was built in under 34 hours for the BIG Hackathon, hosted by the Computer Science Club (CS Club), University of Adelaide and the UniSA Programming Community (UPC), University of South Australia.

The idea is simple:

NexusS helps students merge independent study sessions by connecting them with others nearby.

✨ With NexusS, students can:

🔎 Discover nearby study sessions based on courses or interests
📍 Drop by and join sessions happening around them
📝 Host a session by marking their location and inviting others

NexusS helps university students find study buddies — and maybe even new friends.

MyEvents

Tue, 01 Jul 2025 14:18:10 +0930

Project Overview

Every day, the University of Adelaide and its student clubs host dozens of events-from social mixers and pub crawls to career expos and skill-building workshops. These activities are scattered across Instagram, Facebook and various faculty websites, making it difficult for students-especially first‑years - to stay informed. As a result, many valuable opportunities are missed.

MyEvents is a centralized, university‑focused platform (think Eventbrite for Adelaide Uni) that enables verified students and staff to:

Portfolio

Tue, 01 Jul 2025 14:17:12 +0930

This portfolio was built using Hugo and Paige as an experiment to explore a new tool.