IT Engineering @ 4th Year  ·  Full-Stack Developer  ·  AI / ML  ·  Cybersecurity

 _______________________________________________________________
|| | | ||| | ||| | | ||| | ||| | ||| | | ||| | ||| | | ||| | | ||
||_|_|_|||_|_|||_|_|_|||_|_|||_|_|_|||_|_|||_|_|_|||_|_|||_|_|_||
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|_|


          G H O S T Y   //   ONLINE

   I build things, break things, and then
         figure out why they broke.

I'm Meeth, better known online as Ghostyy.

I'm an IT Engineering student who enjoys building systems where software, intelligence and infrastructure overlap.

My projects tend to fall into a few recurring rabbit holes:

AI / LLMs          → RAG, local models, AI-assisted applications
Cybersecurity      → malware behaviour, anomaly detection, attack simulation
Data & Graphs      → fraud detection, Neo4j, NoSQL systems
Full Stack         → React / Next.js / FastAPI / Python
Experiments        → Minecraft modding, automation, weird side projects

I care less about making another "to-do app" and more about answering:

"What happens if I actually try to build this?"

BASIS SDK A security experimentation and evaluation framework built around attack simulation, behavioural telemetry and ML-based anomaly detection.

Unlike a conventional cybersecurity project that simply classifies pre-existing datasets, BASIS is designed around actively generating controlled attack scenarios against a running system and evaluating whether detection models can actually recognise them.

Python Machine Learning FastAPI Cybersecurity Anomaly Detection Telemetry

Zero-Day Malware Behavior Predictor An AI-assisted malware analysis system aimed at identifying malicious behaviour rather than relying exclusively on known signatures.

The system combines static analysis, behavioural analysis and MITRE ATT&CK mapping to investigate potentially malicious files and generate security-oriented recommendations.

Python Django Machine Learning MITRE ATT&CK Static Analysis Behavioral Analysis

Banking Fraud Detection A graph-oriented approach to financial fraud detection using Neo4j to model relationships between transactions and entities.

The idea is to move beyond evaluating transactions in isolation and instead investigate the suspicious structures that emerge when financial activity is represented as a graph.

Python Neo4j Graph Analytics Fraud Detection NoSQL

Dashboard A full-stack personal dashboard with an integrated AI assistant.

The system separates the frontend and backend into distinct applications and uses local LLM infrastructure to provide AI-powered functionality without making the entire application dependent on external model APIs.

React Tailwind FastAPI Python MongoDB Ollama LLMs

Author An AI-powered writing environment designed for generating long-form content while incorporating external information into the generation pipeline.

React Tailwind CSS FastAPI Python Web Scraping LLMs

Azelia A Discord AI system built around Ollama, bringing locally hosted language models into a conversational environment.

Python Ollama Discord API LLMs

Ghost My Minecraft experimentation ground — a Hypixel SkyBlock QoL mod where I explore game automation, client-side tooling and custom mechanics.

Also, probably the most literal explanation for the username.

Java Minecraft Modding Automation

Alongside the usual web stack, I spend a lot of time experimenting with:

LLMs        → Ollama · Llama · Gemini · LangChain
RAG         → embeddings · vector stores · retrieval pipelines
Databases   → MongoDB · Neo4j · Cassandra · HBase · SQLite
Security    → behavioral analysis · anomaly detection · MITRE ATT&CK

[██████████████████░░]  AI systems
[████████████████░░░░]  Cybersecurity
[███████████████░░░░░]  Full-stack engineering
[██████████████░░░░░░]  Data / Graph systems
[███████████░░░░░░░░░]  Research

Currently interested in the less obvious problems surrounding AI:

• reducing the computational and memory cost of AI systems
• improving how LLM applications retrieve and retain information
• anomaly detection and adversarial behaviour
• graph-based approaches to security and fraud
• building useful AI systems locally rather than treating APIs as magic boxes

I'm increasingly interested in the engineering problems underneath AI rather than just the applications sitting on top of it.

Some areas I'm exploring:

┌─────────────────────────────────────────────────────┐
│  MEMORY                                               │
│  How can AI systems retain useful context without    │
│  continuously increasing the amount of computation? │
├─────────────────────────────────────────────────────┤
│  EFFICIENCY                                           │
│  Can we reduce token usage and inference overhead    │
│  without sacrificing useful information?             │
├─────────────────────────────────────────────────────┤
│  SECURITY                                             │
│  Can models identify malicious behaviour before      │
│  conventional signatures catch it?                   │
├─────────────────────────────────────────────────────┤
│  STRUCTURED KNOWLEDGE                                 │
│  When should an AI system use vectors, graphs,       │
│  databases, or some combination of them?             │
└─────────────────────────────────────────────────────┘

The goal is eventually to turn some of these experiments into proper research rather than leaving them as another repository collecting dust.

If you're interested in AI, security, interesting engineering problems, or just building something unnecessarily complicated:

GitHub Portfolio Email Instagram

        "The ghost isn't the absence of something.

         It's what remains after it leaves."

                              — Ghostyy

_{Built by Meeth / Ghostyy · 2026}