And proved it with math.
Seven arXiv preprints powering the Qualixar platform — behavioral contracts, supply chain security, agent memory, stochastic testing, and multi-agent orchestration.
Varun Pratap Bhardwaj
→ Biologically-inspired forgetting, cognitive quantization, multi-channel retrieval
Read on arXiv →Varun Pratap Bhardwaj
→ 1,980 experiment sessions across 7 models
Read on arXiv →Varun Pratap Bhardwaj
→ 675 tests, 8 novel contributions
Read on arXiv →Varun Pratap Bhardwaj
→ 74.8% on LoCoMo (zero cloud) — highest local-first score reported
Read on arXiv →Varun Pratap Bhardwaj
→ Local-first architecture, Bayesian trust defense against memory poisoning
Read on arXiv →Varun Pratap Bhardwaj
→ Stochastic testing across non-deterministic agent workflows
Read on arXiv →Varun Pratap Bhardwaj
→ Universal Type-C port for AI agents — 25 commands, every transport, every IDE
Read on arXiv →What I investigate.
Agent Behavioral Contracts
Formal specification and runtime enforcement of agent behavior — invariants that hold under all execution paths.
AI Agent Safety & Reliability
Preventing drift, hallucination, and unauthorized actions in autonomous multi-agent systems at production scale.
Supply Chain Security
Detecting, analyzing, and blocking malicious skills before they execute inside agent pipelines.
Formal Methods for AI
Mathematical verification of stochastic, non-deterministic agent systems using contract theory and type-level invariants.
Stochastic Agent Testing
Token-efficient regression frameworks that embrace non-determinism rather than suppressing it.
Eastern Epistemology as Verification
Sāṅkhya and Advaita pramāṇa systems as formal foundations for AI reliability auditing and agent-contract design.