Deepak Mistry

Security Consultant. Seven years breaking apps, networks, and now language models building the open-source tooling I wish I had.

Pentester by trade, researcher by habit. Seven years across application, network, and AI red teaming, with 100+ engagements. Most days I'm trying to understand how AI systems works and break under pressure: agentic misbehaviour, prompt injection, Jailbreaking and what a serious red team should look like when the target isn't deterministic, below are some of my active developments which serve the purpose to test LLMs.

Where I think I add the most value is the build side. I spend a lot of time writing the automation I wish existed during an engagement, and I try to design those tools so the architecture doesn't get in the way later. Having spent real hours inside the current AI stack (serving, agent frameworks, fine-tuning, MCP, Hugging Face, Claude, LM Studio, Ollama, vLLM, llama.cpp,) I can usually pick a tech stack that fits the problem rather than the trend. Still ship open-source steadily, and keep one foot in blockchain and smart-contract work.