Publication record.

Research interests

• Agent faithfulness and causal evaluation
• Grounded tool use and orchestration
• Retrieval and long-term memory
• Failure recovery in multi-step workflows
• Heuristic search and planning
• Privacy-preserving and perspective-aware AI

Peer reviewed

Published in conference proceedings or journals after review.

• Peer reviewed2026

  Breadth-First Search vs. Restarting Random Walks for Escaping Uninformed Heuristic Regions

  D. Platnick, D. Tomasz, E. Earl, S. Khanzadeh, R. Valenzano

  AAAI Conference on Artificial Intelligence

  When a heuristic gives a planner no gradient to follow, the planner stalls. This paper compares breadth-first search against restarting random walks for escaping those regions, and shows the random-walk approach needs only linear worst-case memory.

  My contribution: Co-author on the planner and the experimental evaluation.

  • DOI
  • Case study: Search & planning
  Citation

  @inproceedings{khanzadeh2026restartingrandomwalks,
    title        = {Breadth-First Search vs. Restarting Random Walks for Escaping Uninformed Heuristic Regions},
    author       = {D. Platnick, D. Tomasz, E. Earl, S. Khanzadeh, R. Valenzano},
    booktitle    = {AAAI Conference on Artificial Intelligence},
    year         = {2026},
    url          = {https://doi.org/10.1609/aaai.v40i43.41044},
  }

• Peer reviewed20244 citations (Scholar snapshot)

  GANsemble for Small and Imbalanced Data Sets: A Baseline for Synthetic Microplastics Data

  D. Platnick, S. Khanzadeh, A. Sadeghian, R. A. Valenzano

  Canadian Artificial Intelligence Conference

  A pipeline that pairs an augmentation search with a class-conditional GAN to generate usable synthetic training data from a 210-image imbalanced dataset.

  My contribution: Co-developed the augmentation search and conditional-GAN pipeline.

  • arXiv preprint
  • Proceedings PDF
  • Case study: GANsemble
  Citation

  @inproceedings{khanzadeh2024gansemble,
    title        = {GANsemble for Small and Imbalanced Data Sets: A Baseline for Synthetic Microplastics Data},
    author       = {D. Platnick, S. Khanzadeh, A. Sadeghian, R. A. Valenzano},
    booktitle    = {Canadian Artificial Intelligence Conference},
    year         = {2024},
    url          = {https://arxiv.org/abs/2404.07356},
  }

• Peer reviewed20254 citations (Scholar snapshot)

  An exploratory study on domain knowledge infusion in deep learning for automated threat defense

  S. Khanzadeh, E. C. P. Neto, S. Iqbal, M. Alalfi, S. Buffett

  International Journal of Information Security 24(1), 71

  How injecting domain knowledge into a deep-learning model changes threat detection on large, severely imbalanced telemetry — and where it stops helping.

  My contribution: Lead author on the knowledge-infusion prototype and the experimental evaluation, during the NRC Canada internship.

  • Journal article
  Citation

  @inproceedings{khanzadeh2025domainknowledgeinfusion,
    title        = {An exploratory study on domain knowledge infusion in deep learning for automated threat defense},
    author       = {S. Khanzadeh, E. C. P. Neto, S. Iqbal, M. Alalfi, S. Buffett},
    booktitle    = {International Journal of Information Security 24(1), 71},
    year         = {2025},
    url          = {https://link.springer.com/article/10.1007/s10207-025-00987-4},
  }

• Peer reviewed2025

  Integrating Auxiliary Knowledge into Machine Learning to Improve the Detection of Cyberattacks

  S. Iqbal, S. Khanzadeh, E. C. P. Neto, S. Buffett, M. Sultana, A. Taylor

  International Symposium on Networks, Computers and Communications

  A follow-on study on supplying machine-learning detectors with knowledge they cannot infer from the training data alone.

  • No public link recorded for this entry.
  Citation

  @inproceedings{khanzadeh2025auxiliaryknowledgecyberattacks,
    title        = {Integrating Auxiliary Knowledge into Machine Learning to Improve the Detection of Cyberattacks},
    author       = {S. Iqbal, S. Khanzadeh, E. C. P. Neto, S. Buffett, M. Sultana, A. Taylor},
    booktitle    = {International Symposium on Networks, Computers and Communications},
    year         = {2025},
  }

• Peer reviewed20243 citations (Scholar snapshot)

  Solosphere: A framework for gas optimization in solidity smart contracts

  S. Khanzadeh, M. H. Alalfi

  IEEE International Conference on Software Analysis, Evolution and Reengineering

  A framework that finds and applies gas-cost optimizations in Solidity contracts.

  My contribution: Lead author.

  • IEEE Xplore
  Citation

  @inproceedings{khanzadeh2024solosphere,
    title        = {Solosphere: A framework for gas optimization in solidity smart contracts},
    author       = {S. Khanzadeh, M. H. Alalfi},
    booktitle    = {IEEE International Conference on Software Analysis, Evolution and Reengineering},
    year         = {2024},
    url          = {https://ieeexplore.ieee.org/abstract/document/10621683},
  }

• Peer reviewed202315 citations (Scholar snapshot)

  Optimizing gas consumption in Ethereum smart contracts: Best practices and techniques

  S. Khanzadeh, N. Samreen, M. H. Alalfi

  IEEE International Conference on Software Quality, Reliability and Security

  A survey of the techniques that actually reduce gas consumption in Ethereum smart contracts, and the evidence behind each.

  My contribution: Lead author.

  • IEEE Xplore
  Citation

  @inproceedings{khanzadeh2023gasconsumption,
    title        = {Optimizing gas consumption in Ethereum smart contracts: Best practices and techniques},
    author       = {S. Khanzadeh, N. Samreen, M. H. Alalfi},
    booktitle    = {IEEE International Conference on Software Quality, Reliability and Security},
    year         = {2023},
    url          = {https://ieeexplore.ieee.org/abstract/document/10429984},
  }

Preprints

Public research artifacts that have not been through peer review.

• Preprint2026

  Project Ariadne: A Structural Causal Framework for Auditing Faithfulness in LLM Agents

  S. Khanzadeh

  arXiv

  A protocol for testing whether an LLM agent's stated reasoning actually caused its answer: intervene on the reasoning trace, replay the workflow from the modified state, and check whether the output moves.

  My contribution: Designed the intervention protocol, built the auditing harness, and ran the 30-trajectory evaluation.

  • arXiv preprint
  • Case study: Project Ariadne
  Citation

  @misc{khanzadeh2026projectariadne,
    title        = {Project Ariadne: A Structural Causal Framework for Auditing Faithfulness in LLM Agents},
    author       = {S. Khanzadeh},
    howpublished = {arXiv},
    year         = {2026},
    url          = {https://arxiv.org/abs/2601.02314},
  }

• Preprint20256 citations (Scholar snapshot)

  AgentMesh: A Cooperative Multi-Agent Generative AI Framework for Software Development Automation

  S. Khanzadeh

  arXiv

  A framework in which Planner, Coder, Debugger, and Reviewer agents pass work between each other, using execution feedback to repair their own output rather than stopping at the first failure.

  My contribution: Sole author; designed and implemented the framework.

  • arXiv preprint
  • Case study: AgentMesh
  Citation

  @misc{khanzadeh2025agentmesh,
    title        = {AgentMesh: A Cooperative Multi-Agent Generative AI Framework for Software Development Automation},
    author       = {S. Khanzadeh},
    howpublished = {arXiv},
    year         = {2025},
    url          = {https://arxiv.org/abs/2507.19902},
  }

• Preprint20231 citations (Scholar snapshot)

  Opti Code Pro: A heuristic search-based approach to code refactoring

  S. Khanzadeh, S. A. N. Chan, R. Valenzano, M. Alalfi

  arXiv

  Treats refactoring as a search problem and builds admissible heuristics for it, so the search reaches a target design while expanding far fewer states than uninformed search.

  My contribution: Lead author; designed and implemented the heuristics and the search.

  • arXiv preprint
  • Case study: Search & planning
  Citation

  @misc{khanzadeh2023opticodepro,
    title        = {Opti Code Pro: A heuristic search-based approach to code refactoring},
    author       = {S. Khanzadeh, S. A. N. Chan, R. Valenzano, M. Alalfi},
    howpublished = {arXiv},
    year         = {2023},
    url          = {https://arxiv.org/abs/2305.07594},
  }