Built by Varun Pratap Bhardwaj
Independent Researcher · AI Agent Reliability · ORCID: 0009-0002-8726-4289
The Problem
2.4 million AI agents are deployed today. They make decisions, take actions, and operate autonomously. But when an agent drifts from its intended behavior — recommending the wrong medication, approving a fraudulent transaction, or leaking sensitive data — there are no formal guarantees to catch it.
Monitoring tells you what happened after the damage. agentAssert prevents it before it occurs.
The Solution
agentAssert brings design-by-contract — a proven software engineering discipline — to the world of autonomous AI agents. Combined with stochastic process theory and Lyapunov stability analysis, it provides the first framework with mathematical guarantees that agents stay within specification.
The result: a four-component contract system { P, I, G, R } that defines what agents should do, monitors that they're doing it, and recovers when they don't — with provable drift bounds.
Research Vision
Every autonomous agent should operate under a behavioral contract. Not guidelines. Not suggestions. A formal, enforceable specification with mathematical guarantees. This is how the research community can establish trust in AI systems.
This work is peer-reviewed and published on arXiv (arXiv:2602.22302). The EU AI Act demands runtime evidence by August 2026, and this research provides the theoretical foundations for that requirement.
The Researcher
Varun Pratap Bhardwaj is an independent researcher working at the intersection of formal methods, AI safety, and software engineering. His research focuses on making autonomous AI agents reliable and verifiable through mathematical guarantees.
His published work spans agent behavioral contracts, agent memory systems, and agent security validation, with papers published on arXiv and archived on Zenodo.
The Stability Bowl — our logo represents the Lyapunov stability function V(x) at the heart of agentAssert. The glowing sphere resting in the bowl symbolizes an agent held in stable equilibrium by its behavioral contract. Drift is bounded. Recovery is guaranteed. The agent stays within spec.