PRJ-002Active
Zero-Day Malware Behavior Predictor
Static behavioural analysis, MITRE ATT&CK mapping and an LLM reasoning layer — inferring intent without ever running the file.
- Python
- Django
- Next.js
- TypeScript
- MongoDB
- +3
GhostyyOnline
Meeth W
AI Security Systems Research
I build systems, break them on purpose, and go looking for the part that gave.
Most of my projects start as questions I couldn't leave alone — behavioural security, AI that has to run on my own hardware, and what data looks like when you stop storing it in rows.
- status
- online
- mode
- building
- focus
- ai / security / systems
- location
- Asia, Pacific
ghostyy@localhost: ~
02/ Current focus
What's running right now
One thesis project taking most of the time, and a set of questions running in the background behind it.
Currently buildingBuilding
BASIS SDK
Behavioural anomaly detection that installs as Django middleware, learns what an application's normal traffic looks like, and flags the actors who stop matching it — shipped with the attack scenarios and benchmark harness used to prove it works.
Pipeline
- Attack simulation
- Live application
- Request telemetry
- Feature windows
- ML detection
- Risk score
- Dashboard
- Python
- Django
- scikit-learn
- Isolation Forest
- One-Class SVM
ghostyy@localhost: ~/ps
- basis-sdkrunning
- malware-tracerrunning
- researchrunning
- portfoliobuilding
- sleepnot found
Exploring
- AI memory
- Token efficiency
- RAG systems
- Behavioural cybersecurity
- Graph intelligence
- Local inference
03/ Featured work
Projects worth reading about
Each of these has a case study — what the problem was, what I actually built, and where it falls short.
PRJ-003Research
Graph-Based Fraud Detection
Banking fraud modelled as structure rather than rows — laundering loops and shared-ownership rings surfaced in Neo4j.
- Python
- FastAPI
- Neo4j
- Cypher
- React
- +2
PRJ-004Archived
Personal Dashboard
A self-hosted workspace — notes, files, scraping and an assistant running on a local model rather than someone else's API.
- React
- Vite
- Tailwind CSS
- Python
- FastAPI
- +3
04/ Research.log
Questions I haven't finished with
Not interests — open engineering problems, each with something concrete that would settle it.
[001]AI memory & token efficiencyHow much context does an AI system actually need?[002]Behavioural cybersecurityCan malicious behaviour be detected before a signature exists?[003]Graph intelligenceWhat does fraud look like once transactions become a graph?[004]Retrieval & structured knowledgeWhen should a system reach for vectors, a graph, or an ordinary database?
05/ Stack
What I actually build with
Grouped by what it's for rather than arranged as a wall of logos. Everything here appears in something I've shipped.
01
Languages
Python and TypeScript carry most of the weight; the rest show up where they have to.
- Python
- TypeScript
- JavaScript
- Java
- Kotlin
- C
- C++
02
Frontend
React everywhere, Next.js when routing and rendering matter, Vite when they don't.
- React
- Next.js
- Tailwind CSS
- Vite
- Framer Motion
03
Backend
FastAPI for services, Django when the batteries are the point.
- FastAPI
- Django
- REST APIs
- Uvicorn
04
Databases
Chosen by the shape of the question — documents, graphs, columns or a single file.
- MongoDB
- Neo4j
- Cassandra
- HBase
- SQLite
- Prisma
05
AI / ML
Local models by default; scikit-learn where a classifier beats a language model.
- Ollama
- Llama
- Gemini
- LangChain
- scikit-learn
- PyTorch
- RAG
- Embeddings
06
Infrastructure
Enough to run what I build without renting someone else's opinion about it.
- Linux
- Docker
- Git
- GitHub Actions
- Terraform
06/ GitHub
The public record
This site is hosted from a GitHub repository, so it may as well read from one. Fetched in the browser — there is no server here to hide a token behind, and none is needed.
Languages39 repos
- JavaScript15
- Python13
- TypeScript5
- HTML3
- Java1
- C1
fetching…
Recently pushed
BASIS-SDK
ML-based behavioral anomaly detection as Django security middleware, with a built-in dashboard.
2026-08-12
Python
Aim-Trainer
Browser-based aim training platform with a custom gameplay engine.
2026-06-12
TypeScript
FitAI
ML-powered fitness and health recommendation platform.
2026-04-17
JavaScript
Neural-Networks-Demo
Handwritten digit recognition with a CNN trained on MNIST.
2026-03-29
Python
Malware-Tracer
AI-powered zero-day malware behavior predictor.
2026-02-10
TypeScript
07/ Lab
Not everything needs to become a startup
17 smaller things — a CNN behind a drawing canvas, an API scratchpad, Minecraft mods that taught me more about architecture than any course did, and a bot that fishes.
Open the lab08/ Contact
Say something
Open to conversations about AI systems, security engineering, and problems that don't have a tidy answer yet. No form — a form would need a server, and this site doesn't have one.