Building the verification layer for AI
QWED exists for one question, asked narrowly and answered without hedging: how do you make a model’s output safe to use where correctness, policy, and auditability are not optional? Not safer. Safe enough to act on, and able to show why.
01
Proof over polish
Sounding certain is free. Being right is not. Everything here is built so that an output has to survive a check before it can cross into anything that matters.
02
Safety at the moment of action
AI has stopped talking and started doing. That moves the interesting risk from what a model says to what it is permitted to do — tool use, execution, and what happens when a check fails.
03
A record that outlives the incident
High-stakes deployments get asked, months later, what was checked and why it passed. That question deserves a file, not a recollection.
Who is filing this
A small team. One name on the record so far.
Rahul Dass
Founder
Working on a deterministic verification layer for LLMs and AI agents, so that teams in finance, legal, and infrastructure can use these systems without quietly transferring the risk to whoever reads the output last.
Connect on LinkedInIf you think a system should have to prove things, we are already agreeing about something.
Enterprise deployment, verifiable agent workflows, or the unglamorous work of making AI infrastructure trustworthy — that is the direction this is being built in, and it is a better conversation with more people in it.