Writing Archive

Hi, I’m

Himanshu Mittal

I lead technical work at Solumn AI, where we do AI safety evals for frontier models. I am interested in agent foundations and control research. Previously built industrial optimization systems and a venture-backed hardware-AI startup.

Himanshu Mittal

AI safety ยท Evaluation→

Red Teaming GenAI Safety as a Measurement Problem

A black-box methodology: baseline first, breadth second, depth last. Mapping where a model's safety boundary actually sits instead of collecting jailbreaks.

Faradworks ยท Antler→

Three Questions From Building an AI Hardware Startup

Is the market ready, and what are we building while we wait? Do we replace the workflow or earn our way into it? And what is the AI actually claiming?

Ripik ยท Industrial AI→

Building Optimus: What It Takes to Schedule a Pharmaceutical Plant

What a scheduler has to know about a factory before it can schedule anything, and why I would build the next one around the planner instead of above them.

More writing →