Peer-Reviewed Research

Research

3 peer-reviewed publications. Mathematical foundations for the Agent Software Engineering Suite.

Publications

Formal methods and mathematical guarantees for every layer of the AI agent stack.

arXiv:2602.22302 · Published February 2026

Agent Behavioral Contracts: Formal Specification and Runtime Enforcement for Reliable Autonomous AI Agents

Varun Pratap Bhardwaj

agentAssert · Research Project

We present Agent Behavioral Contracts (ABC), a formal framework that combines design-by-contract principles with stochastic process theory to provide runtime behavioral guarantees for autonomous AI agents. ABC introduces a four-component contract structure {P, I, G, R} with mathematical drift bounds via Lyapunov stability analysis, achieving Θ=0.9541 aggregate compliance across 200 benchmark scenarios.

Read on arXiv Zenodo Archive
Zenodo · 2026

SkillFortify: Formal Verification Framework for Agent Skill Supply Chain Security

Varun Pratap Bhardwaj

SkillFortify · Open Source

A formal verification framework that provides mathematical guarantees for agent skill security across 22 frameworks. Achieves F1=96.95% detection accuracy with zero false positives. SkillFortify addresses the growing supply chain risk in AI agent ecosystems by verifying skill integrity before execution, preventing prompt injection, data exfiltration, and privilege escalation.

Read on Zenodo GitHub PyPI
Zenodo · 2026

SuperLocalMemory: Privacy-Preserving Multi-Agent Memory Architecture

Varun Pratap Bhardwaj

SuperLocalMemory · Open Source

A local-first memory architecture for AI agents that preserves privacy while enabling multi-agent memory sharing. Zero cloud dependency, works with Claude Code, Cursor, ChatGPT, and Perplexity. SuperLocalMemory provides persistent context across sessions with full data sovereignty, ensuring no sensitive information leaves the developer's machine.

Read on Zenodo GitHub npm

Key Results

Rigorously evaluated across diverse enterprise domains.

Θ = 0.9541 Aggregate Compliance Score
200 Benchmark Scenarios
7 Evaluation Domains
97% Drift Recovery Rate

Domain Performance

Benchmark results across 7 real-world enterprise domains.

Code Generation
96.5%
Customer Support
96.2%
Content Moderation
95.8%
Medical Triage
95.1%
Financial Analysis
94.8%
Data Pipeline
94.8%
Legal Research
93.7%

Academic Profiles

Follow ongoing research and publications.

ORCID

0009-0002-8726-4289

View Profile

arXiv Author

Varun Pratap Bhardwaj

View Papers

Research Interests

Building the mathematical and systems foundations for trustworthy autonomous AI.

AI Agent Reliability
Formal Verification
Design-by-Contract
Stochastic Process Theory
Supply Chain Security
Multi-Agent Systems
Privacy-Preserving AI
Runtime Enforcement
EU AI Act Compliance
Enterprise AI Governance

Cite This Work

BibTeX entries for the Agent Software Engineering Suite publications.

BibTeX — Agent Behavioral Contracts 
@article{bhardwaj2026abc,
  title={Agent Behavioral Contracts: Formal Specification and
         Runtime Enforcement for Reliable Autonomous AI Agents},
  author={Bhardwaj, Varun Pratap},
  journal={arXiv preprint arXiv:2602.22302},
  year={2026},
  url={https://arxiv.org/abs/2602.22302},
  doi={10.5281/zenodo.18775393}
}
BibTeX — SkillFortify 
@article{bhardwaj2026skillfortify,
  title={SkillFortify: Formal Verification Framework for
         Agent Skill Supply Chain Security},
  author={Bhardwaj, Varun Pratap},
  year={2026},
  url={https://zenodo.org/records/18787663},
  doi={10.5281/zenodo.18787663}
}
BibTeX — SuperLocalMemory 
@article{bhardwaj2026slm,
  title={SuperLocalMemory: Privacy-Preserving Multi-Agent
         Memory Architecture},
  author={Bhardwaj, Varun Pratap},
  year={2026},
  url={https://zenodo.org/records/18709670},
  doi={10.5281/zenodo.18709670}
}