Writing
Blog
Notes on building things, breaking things, running my own servers and training models at home. Posts are written in Markdown, each one a real page of its own.
-
Building an AI-Powered Threat Intelligence Pipeline
Collect, normalize, analyze, enrich, summarize. Turning the threat-intel firehose into a short digest, with AI as one stage.
-
Hello from the new blog
A short note on why this blog now lives on my own site, and how it works.
-
Designing a Cybersecurity AI Agent Pipeline
Narrow specialized agents, evidence passed between stages, and a fact-check stage that assumes every finding is wrong.
-
LLM Hallucinations in Cybersecurity: Why Verification Matters
Fabricated CVEs, plausible-but-wrong details, and made-up citations. Why every model claim in security work needs verifying.
-
Building the ABC News site on my own
What it takes to build and run a full news site by yourself, from the theme down to the server.
-
Using AI Agents for Cybersecurity: Where They Help and Where They Fail
Where AI agents genuinely help in security work, where they fail with confidence, and how to use them without getting burned.
-
Training a Nepali language model at home
Notes on fine-tuning and serving a Nepali LLM on a single AMD AI Pro 32GB GPU in my homelab.
-
Building a Security Research Assistant with Local LLMs
A private security research helper on local models: RAG for knowledge, tool calling for facts, and keeping output honest.
-
What Happens When an Internet-Facing Server Gets Attacked?
The constant background noise, what the attack surface really is, why logs must leave the box, and a short IR plan.
-
Why Production Engineering Is Different from Writing Code
Writing code that works is the small part. Reliability, monitoring, backups, security and deployment are the rest.
-
From Localhost to Production: Problems Nobody Mentions in Tutorials
DNS, service networking, permissions, containers and env config. The things that break the first few times you ship for real.
-
Why Network Isolation Matters in a Cybersecurity Lab
VLANs, default-deny between segments, and small-scale zero trust. Why isolation keeps a lab from becoming an incident.
-
How I Think About Backups After Designing My Own Infrastructure
Tiers not copies, the target pulls, keep history, test the restore. The backup model I use after building my own infra.
-
What I Learned Designing a Multi-Layer Server Architecture
A VPS, a NAS, a backup box and some tunnels, each with one job. Why the separation happened and what it costs.
-
Cloudflare Tunnels as Part of a Self-Hosted Infrastructure
Exposing self-hosted services without opening ports: what tunnels buy you, how I use them, and where they fall short.
-
Can Multiple AI Agents Work in Parallel? Understanding Parallel Agent Architectures
When running AI agents in parallel actually helps, the architecture that works, and the judging step everyone skips.
-
Why Nepali Is Still a Hard Language for AI Models
Thin data, tokenizers built for English, and manual evaluation. The concrete reasons Nepali still lags, and what helps.
-
Running Large Language Models Locally: What VRAM Actually Means
Parameters are not gigabytes. Where VRAM really goes, why quantization works, and why a 70B model does not need 70GB.
No posts yet. Check back soon.