Jyoti Ranjan
  • Home
  • Projects
  • Blog
  • About
  • Contact
Jyoti Ranjan
GenAI Architect · Applied LLM Systems

Jyoti Ranjan

I build production-grade AI systems — and, increasingly, the security and evidence layer they need before anyone can trust them in a regulated setting. Nineteen years in the field, a US patent, and an arXiv paper along the way.

Read the blog · Projects · About · GitHub

Lately I’ve been working at the intersection of agentic AI and security — how autonomous agents can be deployed safely, and how you prove to an auditor that they are. That work shows up here as writing, open-source, and the occasional deep-dive.

Writing

Your MCP Server Is an Attack Surface

MCP
AI security
agents
compliance

MCP let us give AI agents hands. It also gave attackers a new, under-audited surface: the tool descriptions and schemas an agent reads before it acts. We trace each of the six weaknesses down to the offending bytes, then look at the gap between spotting them and proving to an auditor that you did.

Jul 4, 2026

Understanding LLM Parameters: A Step-by-Step Calculation Guide Using GPT-2

LLM
Deep Learning
GPT-2
Transformers

A comprehensive breakdown of how to calculate the total number of parameters in GPT-2, from input embeddings to output predictions.

Nov 21, 2024
No matching items