cd ~/about
Meeth W
Known online as Ghostyy. Final-year IT engineering student, and someone who has never been able to leave a system alone once it started behaving strangely.
- Alias
- ghostyy
- Location
- Asia, Pacific
I'm in my final year of a B.E. in Information Technology. Most of what I actually know came from building things slightly beyond what I could justify at the time, then having to understand them well enough to fix what broke.
The through-line across my work is a distrust of black boxes. Almost everything I build with language models runs locally, because a hosted API hides exactly the constraints — latency, memory, context cost — that determine how a system should be designed. The same instinct shows up in security: I would rather generate the attack, watch the telemetry, and find out whether detection holds than read a benchmark that says it does.
Right now most of my time goes to BASIS SDK, my engineering thesis — behavioural anomaly detection shipped as Django middleware, along with the experiment suite that measures it honestly enough to report its own blind spots.
I'm not especially interested in building another CRUD application. Most of my projects start as questions I couldn't leave alone, and the good ones end with a clearer understanding of why the obvious approach doesn't work.
whoami
meeth W · ghostyy
cat /etc/education
B.E. Information Technology
Mumbai University
Final year
groups
ai security systems data research
The ghost isn't the absence of something. It's what remains after it leaves.
Where my attention goes
Six areas that keep showing up in what I choose to build.
AI systems
Local models, retrieval, and the engineering underneath rather than the demo on top.
Cybersecurity
Behavioural detection, anomaly scoring, and attacks that don't have a signature yet.
Full-stack systems
Whole applications — API, storage, interface — because the interesting bugs live at the seams.
Graph systems
Structure as the thing being queried, not as an afterthought bolted onto rows.
Data engineering
Picking a store by the shape of the question instead of by habit.
Research
Turning the recurring questions into something measurable rather than another repository.
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.
Languages
Python and TypeScript carry most of the weight; the rest show up where they have to.
- Python
- TypeScript
- JavaScript
- Java
- Kotlin
- C
- C++
Frontend
React everywhere, Next.js when routing and rendering matter, Vite when they don't.
- React
- Next.js
- Tailwind CSS
- Vite
- Framer Motion
Backend
FastAPI for services, Django when the batteries are the point.
- FastAPI
- Django
- REST APIs
- Uvicorn
Databases
Chosen by the shape of the question — documents, graphs, columns or a single file.
- MongoDB
- Neo4j
- Cassandra
- HBase
- SQLite
- Prisma
AI / ML
Local models by default; scikit-learn where a classifier beats a language model.
- Ollama
- Llama
- Gemini
- LangChain
- scikit-learn
- PyTorch
- RAG
- Embeddings
Infrastructure
Enough to run what I build without renting someone else's opinion about it.
- Linux
- Docker
- Git
- GitHub Actions
- Terraform
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.