WALL-E
In trainingAn 80M-parameter English language model with no pretrained weights and no paid GPUs — tokenizer, architecture, and training loop all hand-built to understand the whole stack.
Hariharan V · Gen AI Engineer
I build AI agents that do real work — including the kind that breaks in.
Gen AI engineer working where language models meet security. By day I'm building an autonomous penetration-testing agent; after hours I'm training an 80-million-parameter language model from scratch — tokenizer to training loop — on free GPUs. I like small budgets and sharp tools.
// currently: LLM agents × offensive security
Selected work
Projects
Pentesting agent
Day jobAn autonomous agent that finds vulnerabilities, confirms them, and asks a human before exploiting anything. Writes reports engineers and executives can both read.
haricane
You are hereThis site — and the private, API-first workspace behind it. Cloudflare Pages + D1, designed so a future agent can manage my projects and notes through the same API I use.
Saga Weaver
MVPA reading companion for webnovels with a spoiler-safe knowledge base — it only knows what you've read up to your current chapter.
Tag, multiplayer
Weekend buildReal-time 2D tag in the browser — Durable Objects for rooms, WebSockets for sync, Canvas for the chase.
Current intensity
Skills
Rated on the Saffir–Simpson scale — Cat 1 learning it, Cat 3 comfortable shipping, Cat 5 deep water. The name was right there.
AI / LLMs
- Large language models Cat 5
- Agentic systems Cat 5
- Tokenization & small models Cat 3
Security
- Web app security Cat 1
- Pentesting fundamentals Cat 1
Infra / Tools
- Cloudflare (Pages, Workers, D1) Cat 3
- Python Cat 3
- Git / CI Cat 1
Make landfall
Contact
Found this through my resume? Email reaches me fastest.